Upload 220 to 300 steps of 351 steps.

checkpoint-220/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ../ckpts/Meta-Llama-3-8B-Instruct
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.11.1

checkpoint-220/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "../ckpts/Meta-Llama-3-8B-Instruct",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "gate_proj",
+    "q_proj",
+    "lm_head",
+    "o_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-220/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c413d5a5e805e1f58ba270fcdc8af63e5253b13148b768003336b65d47375fb3
+size 1138856856

checkpoint-220/global_step220/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56eb33ef90047b90033906762ef15e399b1c33901bd6b6892da3e1a287cca3a2
+size 528781328

checkpoint-220/global_step220/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4aa299524e2641ee382b7d61759463ce8bcfecc5cbe9ed4ecd043623206da055
+size 199905337

checkpoint-220/latest

@@ -0,0 +1 @@
+global_step220

checkpoint-220/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c48894345a22525b1e69a19e3f3dc55a1aeef79df93012160905d737bdd2a920
+size 14244

checkpoint-220/trainer_state.json

@@ -0,0 +1,1573 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 1.8773333333333333,
+  "eval_steps": 500,
+  "global_step": 220,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008533333333333334,
+      "grad_norm": 100.21848203113535,
+      "learning_rate": 0.0,
+      "loss": 7.1962,
+      "step": 1
+    },
+    {
+      "epoch": 0.017066666666666667,
+      "grad_norm": 96.43006188910957,
+      "learning_rate": 3.0102999566398115e-06,
+      "loss": 6.9414,
+      "step": 2
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 97.35803466618715,
+      "learning_rate": 4.771212547196624e-06,
+      "loss": 7.0102,
+      "step": 3
+    },
+    {
+      "epoch": 0.034133333333333335,
+      "grad_norm": 95.14837816372646,
+      "learning_rate": 6.020599913279623e-06,
+      "loss": 6.5295,
+      "step": 4
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 91.76544275692784,
+      "learning_rate": 6.989700043360187e-06,
+      "loss": 6.4806,
+      "step": 5
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 84.4494318688335,
+      "learning_rate": 7.781512503836437e-06,
+      "loss": 6.4194,
+      "step": 6
+    },
+    {
+      "epoch": 0.05973333333333333,
+      "grad_norm": 71.37977490595638,
+      "learning_rate": 8.450980400142568e-06,
+      "loss": 5.4953,
+      "step": 7
+    },
+    {
+      "epoch": 0.06826666666666667,
+      "grad_norm": 49.31153456754566,
+      "learning_rate": 9.030899869919434e-06,
+      "loss": 5.4123,
+      "step": 8
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 20.37296364560341,
+      "learning_rate": 9.542425094393249e-06,
+      "loss": 5.2334,
+      "step": 9
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 7.968467079076881,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 5.0282,
+      "step": 10
+    },
+    {
+      "epoch": 0.09386666666666667,
+      "grad_norm": 3.559446532055649,
+      "learning_rate": 1.041392685158225e-05,
+      "loss": 4.612,
+      "step": 11
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 3.5528846947995674,
+      "learning_rate": 1.0791812460476248e-05,
+      "loss": 4.9475,
+      "step": 12
+    },
+    {
+      "epoch": 0.11093333333333333,
+      "grad_norm": 3.541968897471334,
+      "learning_rate": 1.1139433523068365e-05,
+      "loss": 4.2777,
+      "step": 13
+    },
+    {
+      "epoch": 0.11946666666666667,
+      "grad_norm": 3.54718070036198,
+      "learning_rate": 1.1461280356782378e-05,
+      "loss": 4.3507,
+      "step": 14
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 3.8632334830606747,
+      "learning_rate": 1.1760912590556813e-05,
+      "loss": 4.5364,
+      "step": 15
+    },
+    {
+      "epoch": 0.13653333333333334,
+      "grad_norm": 3.6637424744054004,
+      "learning_rate": 1.2041199826559246e-05,
+      "loss": 3.9672,
+      "step": 16
+    },
+    {
+      "epoch": 0.14506666666666668,
+      "grad_norm": 3.919802904818311,
+      "learning_rate": 1.230448921378274e-05,
+      "loss": 4.0618,
+      "step": 17
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 4.71904950738746,
+      "learning_rate": 1.2552725051033058e-05,
+      "loss": 4.6656,
+      "step": 18
+    },
+    {
+      "epoch": 0.16213333333333332,
+      "grad_norm": 4.6656317698690835,
+      "learning_rate": 1.2787536009528288e-05,
+      "loss": 4.1131,
+      "step": 19
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 5.145138692367417,
+      "learning_rate": 1.301029995663981e-05,
+      "loss": 4.0989,
+      "step": 20
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 5.923538014759818,
+      "learning_rate": 1.3222192947339192e-05,
+      "loss": 4.4991,
+      "step": 21
+    },
+    {
+      "epoch": 0.18773333333333334,
+      "grad_norm": 5.941056962941364,
+      "learning_rate": 1.3424226808222062e-05,
+      "loss": 4.0836,
+      "step": 22
+    },
+    {
+      "epoch": 0.19626666666666667,
+      "grad_norm": 6.171026012117947,
+      "learning_rate": 1.3617278360175927e-05,
+      "loss": 3.6861,
+      "step": 23
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 7.130542138930838,
+      "learning_rate": 1.380211241711606e-05,
+      "loss": 4.0958,
+      "step": 24
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 7.328837606110418,
+      "learning_rate": 1.3979400086720374e-05,
+      "loss": 3.9524,
+      "step": 25
+    },
+    {
+      "epoch": 0.22186666666666666,
+      "grad_norm": 6.923489005711429,
+      "learning_rate": 1.4149733479708178e-05,
+      "loss": 3.6062,
+      "step": 26
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 6.409498926059221,
+      "learning_rate": 1.4313637641589872e-05,
+      "loss": 3.2034,
+      "step": 27
+    },
+    {
+      "epoch": 0.23893333333333333,
+      "grad_norm": 5.781628405584682,
+      "learning_rate": 1.4471580313422191e-05,
+      "loss": 2.8158,
+      "step": 28
+    },
+    {
+      "epoch": 0.24746666666666667,
+      "grad_norm": 6.2927590068243315,
+      "learning_rate": 1.4623979978989559e-05,
+      "loss": 2.9803,
+      "step": 29
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 6.103844678630006,
+      "learning_rate": 1.4771212547196623e-05,
+      "loss": 2.847,
+      "step": 30
+    },
+    {
+      "epoch": 0.26453333333333334,
+      "grad_norm": 7.656341925867144,
+      "learning_rate": 1.4913616938342726e-05,
+      "loss": 3.0907,
+      "step": 31
+    },
+    {
+      "epoch": 0.2730666666666667,
+      "grad_norm": 6.324242877501844,
+      "learning_rate": 1.5051499783199059e-05,
+      "loss": 2.3467,
+      "step": 32
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 7.606313101162739,
+      "learning_rate": 1.5185139398778874e-05,
+      "loss": 2.5292,
+      "step": 33
+    },
+    {
+      "epoch": 0.29013333333333335,
+      "grad_norm": 8.553792493849265,
+      "learning_rate": 1.531478917042255e-05,
+      "loss": 2.4547,
+      "step": 34
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 8.483368703272543,
+      "learning_rate": 1.5440680443502753e-05,
+      "loss": 2.1956,
+      "step": 35
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 8.364739352838077,
+      "learning_rate": 1.5563025007672873e-05,
+      "loss": 1.8552,
+      "step": 36
+    },
+    {
+      "epoch": 0.3157333333333333,
+      "grad_norm": 9.37663682000104,
+      "learning_rate": 1.5682017240669948e-05,
+      "loss": 1.9228,
+      "step": 37
+    },
+    {
+      "epoch": 0.32426666666666665,
+      "grad_norm": 9.823047193440066,
+      "learning_rate": 1.57978359661681e-05,
+      "loss": 1.7033,
+      "step": 38
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 9.692618512955894,
+      "learning_rate": 1.591064607026499e-05,
+      "loss": 1.3768,
+      "step": 39
+    },
+    {
+      "epoch": 0.3413333333333333,
+      "grad_norm": 9.08889021911031,
+      "learning_rate": 1.6020599913279622e-05,
+      "loss": 1.3015,
+      "step": 40
+    },
+    {
+      "epoch": 0.34986666666666666,
+      "grad_norm": 8.081534221516058,
+      "learning_rate": 1.6127838567197353e-05,
+      "loss": 0.9228,
+      "step": 41
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 6.238638048950311,
+      "learning_rate": 1.6232492903979005e-05,
+      "loss": 0.7267,
+      "step": 42
+    },
+    {
+      "epoch": 0.36693333333333333,
+      "grad_norm": 3.4058036861773604,
+      "learning_rate": 1.6334684555795865e-05,
+      "loss": 0.5875,
+      "step": 43
+    },
+    {
+      "epoch": 0.37546666666666667,
+      "grad_norm": 2.079163829467713,
+      "learning_rate": 1.6434526764861872e-05,
+      "loss": 0.6355,
+      "step": 44
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 1.5597487833024746,
+      "learning_rate": 1.6532125137753435e-05,
+      "loss": 0.5106,
+      "step": 45
+    },
+    {
+      "epoch": 0.39253333333333335,
+      "grad_norm": 2.491689602375256,
+      "learning_rate": 1.662757831681574e-05,
+      "loss": 0.6454,
+      "step": 46
+    },
+    {
+      "epoch": 0.4010666666666667,
+      "grad_norm": 2.010880438195854,
+      "learning_rate": 1.672097857935717e-05,
+      "loss": 0.4757,
+      "step": 47
+    },
+    {
+      "epoch": 0.4096,
+      "grad_norm": 1.9452805114322096,
+      "learning_rate": 1.681241237375587e-05,
+      "loss": 0.4133,
+      "step": 48
+    },
+    {
+      "epoch": 0.41813333333333336,
+      "grad_norm": 1.7620848552306103,
+      "learning_rate": 1.6901960800285137e-05,
+      "loss": 0.4004,
+      "step": 49
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 1.278224489774809,
+      "learning_rate": 1.6989700043360187e-05,
+      "loss": 0.3523,
+      "step": 50
+    },
+    {
+      "epoch": 0.4352,
+      "grad_norm": 1.6151354758303231,
+      "learning_rate": 1.7075701760979363e-05,
+      "loss": 0.4317,
+      "step": 51
+    },
+    {
+      "epoch": 0.4437333333333333,
+      "grad_norm": 1.3451396055695035,
+      "learning_rate": 1.716003343634799e-05,
+      "loss": 0.3474,
+      "step": 52
+    },
+    {
+      "epoch": 0.45226666666666665,
+      "grad_norm": 1.6814977782362666,
+      "learning_rate": 1.724275869600789e-05,
+      "loss": 0.3706,
+      "step": 53
+    },
+    {
+      "epoch": 0.4608,
+      "grad_norm": 1.1682442432688667,
+      "learning_rate": 1.7323937598229687e-05,
+      "loss": 0.3488,
+      "step": 54
+    },
+    {
+      "epoch": 0.4693333333333333,
+      "grad_norm": 0.8839814540462471,
+      "learning_rate": 1.7403626894942437e-05,
+      "loss": 0.293,
+      "step": 55
+    },
+    {
+      "epoch": 0.47786666666666666,
+      "grad_norm": 0.7974304806295485,
+      "learning_rate": 1.7481880270062003e-05,
+      "loss": 0.2717,
+      "step": 56
+    },
+    {
+      "epoch": 0.4864,
+      "grad_norm": 0.6232203657585239,
+      "learning_rate": 1.7558748556724913e-05,
+      "loss": 0.1741,
+      "step": 57
+    },
+    {
+      "epoch": 0.49493333333333334,
+      "grad_norm": 0.6850688604563008,
+      "learning_rate": 1.763427993562937e-05,
+      "loss": 0.228,
+      "step": 58
+    },
+    {
+      "epoch": 0.5034666666666666,
+      "grad_norm": 0.5923826384300431,
+      "learning_rate": 1.7708520116421443e-05,
+      "loss": 0.2131,
+      "step": 59
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.489754430485032,
+      "learning_rate": 1.7781512503836432e-05,
+      "loss": 0.165,
+      "step": 60
+    },
+    {
+      "epoch": 0.5205333333333333,
+      "grad_norm": 0.5280940052395061,
+      "learning_rate": 1.7853298350107667e-05,
+      "loss": 0.1658,
+      "step": 61
+    },
+    {
+      "epoch": 0.5290666666666667,
+      "grad_norm": 0.4750905992036739,
+      "learning_rate": 1.7923916894982537e-05,
+      "loss": 0.1438,
+      "step": 62
+    },
+    {
+      "epoch": 0.5376,
+      "grad_norm": 0.4162998392722401,
+      "learning_rate": 1.7993405494535815e-05,
+      "loss": 0.1555,
+      "step": 63
+    },
+    {
+      "epoch": 0.5461333333333334,
+      "grad_norm": 0.26863266246370443,
+      "learning_rate": 1.806179973983887e-05,
+      "loss": 0.1323,
+      "step": 64
+    },
+    {
+      "epoch": 0.5546666666666666,
+      "grad_norm": 0.26534221125601215,
+      "learning_rate": 1.8129133566428553e-05,
+      "loss": 0.1671,
+      "step": 65
+    },
+    {
+      "epoch": 0.5632,
+      "grad_norm": 0.2548662962257576,
+      "learning_rate": 1.8195439355418686e-05,
+      "loss": 0.1308,
+      "step": 66
+    },
+    {
+      "epoch": 0.5717333333333333,
+      "grad_norm": 0.18045394638578796,
+      "learning_rate": 1.8260748027008263e-05,
+      "loss": 0.1262,
+      "step": 67
+    },
+    {
+      "epoch": 0.5802666666666667,
+      "grad_norm": 0.17070388073612064,
+      "learning_rate": 1.8325089127062364e-05,
+      "loss": 0.1192,
+      "step": 68
+    },
+    {
+      "epoch": 0.5888,
+      "grad_norm": 0.1531381679244776,
+      "learning_rate": 1.8388490907372553e-05,
+      "loss": 0.1274,
+      "step": 69
+    },
+    {
+      "epoch": 0.5973333333333334,
+      "grad_norm": 0.18196434993687946,
+      "learning_rate": 1.8450980400142568e-05,
+      "loss": 0.1375,
+      "step": 70
+    },
+    {
+      "epoch": 0.6058666666666667,
+      "grad_norm": 0.15324416972951205,
+      "learning_rate": 1.8512583487190752e-05,
+      "loss": 0.1599,
+      "step": 71
+    },
+    {
+      "epoch": 0.6144,
+      "grad_norm": 0.10884462064503801,
+      "learning_rate": 1.857332496431268e-05,
+      "loss": 0.1041,
+      "step": 72
+    },
+    {
+      "epoch": 0.6229333333333333,
+      "grad_norm": 0.12915133528192668,
+      "learning_rate": 1.8633228601204555e-05,
+      "loss": 0.1406,
+      "step": 73
+    },
+    {
+      "epoch": 0.6314666666666666,
+      "grad_norm": 0.12553425699952878,
+      "learning_rate": 1.8692317197309763e-05,
+      "loss": 0.1256,
+      "step": 74
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11976960918968543,
+      "learning_rate": 1.8750612633916997e-05,
+      "loss": 0.1144,
+      "step": 75
+    },
+    {
+      "epoch": 0.6485333333333333,
+      "grad_norm": 0.115805998298789,
+      "learning_rate": 1.8808135922807914e-05,
+      "loss": 0.1528,
+      "step": 76
+    },
+    {
+      "epoch": 0.6570666666666667,
+      "grad_norm": 0.10325948496697443,
+      "learning_rate": 1.8864907251724818e-05,
+      "loss": 0.1044,
+      "step": 77
+    },
+    {
+      "epoch": 0.6656,
+      "grad_norm": 0.09595064346541006,
+      "learning_rate": 1.8920946026904802e-05,
+      "loss": 0.1534,
+      "step": 78
+    },
+    {
+      "epoch": 0.6741333333333334,
+      "grad_norm": 0.08796742845240496,
+      "learning_rate": 1.8976270912904414e-05,
+      "loss": 0.1155,
+      "step": 79
+    },
+    {
+      "epoch": 0.6826666666666666,
+      "grad_norm": 0.08218991738379527,
+      "learning_rate": 1.9030899869919434e-05,
+      "loss": 0.1311,
+      "step": 80
+    },
+    {
+      "epoch": 0.6912,
+      "grad_norm": 0.08290815261109215,
+      "learning_rate": 1.9084850188786497e-05,
+      "loss": 0.11,
+      "step": 81
+    },
+    {
+      "epoch": 0.6997333333333333,
+      "grad_norm": 0.0794132180176064,
+      "learning_rate": 1.9138138523837165e-05,
+      "loss": 0.1135,
+      "step": 82
+    },
+    {
+      "epoch": 0.7082666666666667,
+      "grad_norm": 0.06934410705255296,
+      "learning_rate": 1.919078092376074e-05,
+      "loss": 0.109,
+      "step": 83
+    },
+    {
+      "epoch": 0.7168,
+      "grad_norm": 0.09000563031870593,
+      "learning_rate": 1.9242792860618813e-05,
+      "loss": 0.12,
+      "step": 84
+    },
+    {
+      "epoch": 0.7253333333333334,
+      "grad_norm": 0.1134042277879818,
+      "learning_rate": 1.929418925714293e-05,
+      "loss": 0.1223,
+      "step": 85
+    },
+    {
+      "epoch": 0.7338666666666667,
+      "grad_norm": 0.09118764690233076,
+      "learning_rate": 1.9344984512435673e-05,
+      "loss": 0.1459,
+      "step": 86
+    },
+    {
+      "epoch": 0.7424,
+      "grad_norm": 0.07873016754353963,
+      "learning_rate": 1.9395192526186183e-05,
+      "loss": 0.1422,
+      "step": 87
+    },
+    {
+      "epoch": 0.7509333333333333,
+      "grad_norm": 0.1796495874463076,
+      "learning_rate": 1.9444826721501687e-05,
+      "loss": 0.1291,
+      "step": 88
+    },
+    {
+      "epoch": 0.7594666666666666,
+      "grad_norm": 0.0679589944174269,
+      "learning_rate": 1.9493900066449125e-05,
+      "loss": 0.108,
+      "step": 89
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.08174688574235538,
+      "learning_rate": 1.9542425094393246e-05,
+      "loss": 0.1081,
+      "step": 90
+    },
+    {
+      "epoch": 0.7765333333333333,
+      "grad_norm": 0.057137370501406756,
+      "learning_rate": 1.9590413923210934e-05,
+      "loss": 0.0934,
+      "step": 91
+    },
+    {
+      "epoch": 0.7850666666666667,
+      "grad_norm": 0.06578111924908255,
+      "learning_rate": 1.9637878273455555e-05,
+      "loss": 0.1085,
+      "step": 92
+    },
+    {
+      "epoch": 0.7936,
+      "grad_norm": 0.08945990540906254,
+      "learning_rate": 1.968482948553935e-05,
+      "loss": 0.1747,
+      "step": 93
+    },
+    {
+      "epoch": 0.8021333333333334,
+      "grad_norm": 0.06183863311044229,
+      "learning_rate": 1.9731278535996986e-05,
+      "loss": 0.1136,
+      "step": 94
+    },
+    {
+      "epoch": 0.8106666666666666,
+      "grad_norm": 0.05777899602544702,
+      "learning_rate": 1.9777236052888476e-05,
+      "loss": 0.0984,
+      "step": 95
+    },
+    {
+      "epoch": 0.8192,
+      "grad_norm": 0.08130851607693534,
+      "learning_rate": 1.9822712330395683e-05,
+      "loss": 0.187,
+      "step": 96
+    },
+    {
+      "epoch": 0.8277333333333333,
+      "grad_norm": 0.06426546202002927,
+      "learning_rate": 1.986771734266245e-05,
+      "loss": 0.1296,
+      "step": 97
+    },
+    {
+      "epoch": 0.8362666666666667,
+      "grad_norm": 0.069692313707994,
+      "learning_rate": 1.991226075692495e-05,
+      "loss": 0.1404,
+      "step": 98
+    },
+    {
+      "epoch": 0.8448,
+      "grad_norm": 0.05494542266886729,
+      "learning_rate": 1.9956351945975496e-05,
+      "loss": 0.116,
+      "step": 99
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.07571686966840627,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 0.1539,
+      "step": 100
+    },
+    {
+      "epoch": 0.8618666666666667,
+      "grad_norm": 0.054351059117603705,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 101
+    },
+    {
+      "epoch": 0.8704,
+      "grad_norm": 0.06531899611551092,
+      "learning_rate": 2e-05,
+      "loss": 0.0827,
+      "step": 102
+    },
+    {
+      "epoch": 0.8789333333333333,
+      "grad_norm": 0.06131678646504652,
+      "learning_rate": 2e-05,
+      "loss": 0.1266,
+      "step": 103
+    },
+    {
+      "epoch": 0.8874666666666666,
+      "grad_norm": 0.06850220540661824,
+      "learning_rate": 2e-05,
+      "loss": 0.1456,
+      "step": 104
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.05806908951252483,
+      "learning_rate": 2e-05,
+      "loss": 0.0954,
+      "step": 105
+    },
+    {
+      "epoch": 0.9045333333333333,
+      "grad_norm": 0.06503642452033717,
+      "learning_rate": 2e-05,
+      "loss": 0.1417,
+      "step": 106
+    },
+    {
+      "epoch": 0.9130666666666667,
+      "grad_norm": 0.050486271853277066,
+      "learning_rate": 2e-05,
+      "loss": 0.0959,
+      "step": 107
+    },
+    {
+      "epoch": 0.9216,
+      "grad_norm": 0.07746063813802379,
+      "learning_rate": 2e-05,
+      "loss": 0.1256,
+      "step": 108
+    },
+    {
+      "epoch": 0.9301333333333334,
+      "grad_norm": 0.051231172380840004,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 109
+    },
+    {
+      "epoch": 0.9386666666666666,
+      "grad_norm": 0.056296443557859455,
+      "learning_rate": 2e-05,
+      "loss": 0.1056,
+      "step": 110
+    },
+    {
+      "epoch": 0.9472,
+      "grad_norm": 0.05058663240713958,
+      "learning_rate": 2e-05,
+      "loss": 0.0971,
+      "step": 111
+    },
+    {
+      "epoch": 0.9557333333333333,
+      "grad_norm": 0.05532886570130611,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 112
+    },
+    {
+      "epoch": 0.9642666666666667,
+      "grad_norm": 0.05327811326654907,
+      "learning_rate": 2e-05,
+      "loss": 0.0989,
+      "step": 113
+    },
+    {
+      "epoch": 0.9728,
+      "grad_norm": 0.05663279364147864,
+      "learning_rate": 2e-05,
+      "loss": 0.0958,
+      "step": 114
+    },
+    {
+      "epoch": 0.9813333333333333,
+      "grad_norm": 0.04930904541225805,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 115
+    },
+    {
+      "epoch": 0.9898666666666667,
+      "grad_norm": 0.06096947951115022,
+      "learning_rate": 2e-05,
+      "loss": 0.106,
+      "step": 116
+    },
+    {
+      "epoch": 0.9984,
+      "grad_norm": 0.050092322361182495,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 117
+    },
+    {
+      "epoch": 1.0069333333333332,
+      "grad_norm": 0.04980408443758999,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 118
+    },
+    {
+      "epoch": 1.0154666666666667,
+      "grad_norm": 0.051183082721834305,
+      "learning_rate": 2e-05,
+      "loss": 0.1049,
+      "step": 119
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.04332220265802814,
+      "learning_rate": 2e-05,
+      "loss": 0.0983,
+      "step": 120
+    },
+    {
+      "epoch": 1.0325333333333333,
+      "grad_norm": 0.07211731499677299,
+      "learning_rate": 2e-05,
+      "loss": 0.1386,
+      "step": 121
+    },
+    {
+      "epoch": 1.0410666666666666,
+      "grad_norm": 0.06550870223740553,
+      "learning_rate": 2e-05,
+      "loss": 0.1334,
+      "step": 122
+    },
+    {
+      "epoch": 1.0496,
+      "grad_norm": 0.05331839690767287,
+      "learning_rate": 2e-05,
+      "loss": 0.1014,
+      "step": 123
+    },
+    {
+      "epoch": 1.0581333333333334,
+      "grad_norm": 0.05227685628767905,
+      "learning_rate": 2e-05,
+      "loss": 0.1098,
+      "step": 124
+    },
+    {
+      "epoch": 1.0666666666666667,
+      "grad_norm": 0.07641693882491171,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 125
+    },
+    {
+      "epoch": 1.0752,
+      "grad_norm": 0.052835367770791786,
+      "learning_rate": 2e-05,
+      "loss": 0.1231,
+      "step": 126
+    },
+    {
+      "epoch": 1.0837333333333334,
+      "grad_norm": 0.07520329755025788,
+      "learning_rate": 2e-05,
+      "loss": 0.085,
+      "step": 127
+    },
+    {
+      "epoch": 1.0922666666666667,
+      "grad_norm": 0.07670066152157425,
+      "learning_rate": 2e-05,
+      "loss": 0.1071,
+      "step": 128
+    },
+    {
+      "epoch": 1.1008,
+      "grad_norm": 0.052832906560645154,
+      "learning_rate": 2e-05,
+      "loss": 0.1093,
+      "step": 129
+    },
+    {
+      "epoch": 1.1093333333333333,
+      "grad_norm": 0.06573889037311398,
+      "learning_rate": 2e-05,
+      "loss": 0.1193,
+      "step": 130
+    },
+    {
+      "epoch": 1.1178666666666666,
+      "grad_norm": 0.05175471296566334,
+      "learning_rate": 2e-05,
+      "loss": 0.1184,
+      "step": 131
+    },
+    {
+      "epoch": 1.1264,
+      "grad_norm": 0.05912231419793496,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 132
+    },
+    {
+      "epoch": 1.1349333333333333,
+      "grad_norm": 0.04899140475981105,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 133
+    },
+    {
+      "epoch": 1.1434666666666666,
+      "grad_norm": 0.05939916939142137,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 134
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.0516819217599706,
+      "learning_rate": 2e-05,
+      "loss": 0.0834,
+      "step": 135
+    },
+    {
+      "epoch": 1.1605333333333334,
+      "grad_norm": 0.05456440346454737,
+      "learning_rate": 2e-05,
+      "loss": 0.1183,
+      "step": 136
+    },
+    {
+      "epoch": 1.1690666666666667,
+      "grad_norm": 0.059906464476343235,
+      "learning_rate": 2e-05,
+      "loss": 0.1048,
+      "step": 137
+    },
+    {
+      "epoch": 1.1776,
+      "grad_norm": 0.0720112680204319,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 138
+    },
+    {
+      "epoch": 1.1861333333333333,
+      "grad_norm": 0.04940202805828527,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 139
+    },
+    {
+      "epoch": 1.1946666666666665,
+      "grad_norm": 0.060088609545130046,
+      "learning_rate": 2e-05,
+      "loss": 0.0952,
+      "step": 140
+    },
+    {
+      "epoch": 1.2032,
+      "grad_norm": 0.04694761423612446,
+      "learning_rate": 2e-05,
+      "loss": 0.0717,
+      "step": 141
+    },
+    {
+      "epoch": 1.2117333333333333,
+      "grad_norm": 0.05628581562512457,
+      "learning_rate": 2e-05,
+      "loss": 0.1062,
+      "step": 142
+    },
+    {
+      "epoch": 1.2202666666666666,
+      "grad_norm": 0.06876420990437652,
+      "learning_rate": 2e-05,
+      "loss": 0.1218,
+      "step": 143
+    },
+    {
+      "epoch": 1.2288000000000001,
+      "grad_norm": 0.058774700501610655,
+      "learning_rate": 2e-05,
+      "loss": 0.1125,
+      "step": 144
+    },
+    {
+      "epoch": 1.2373333333333334,
+      "grad_norm": 0.061855922064341186,
+      "learning_rate": 2e-05,
+      "loss": 0.1295,
+      "step": 145
+    },
+    {
+      "epoch": 1.2458666666666667,
+      "grad_norm": 0.0813047704730138,
+      "learning_rate": 2e-05,
+      "loss": 0.1165,
+      "step": 146
+    },
+    {
+      "epoch": 1.2544,
+      "grad_norm": 0.061374000305305446,
+      "learning_rate": 2e-05,
+      "loss": 0.1094,
+      "step": 147
+    },
+    {
+      "epoch": 1.2629333333333332,
+      "grad_norm": 0.055537169110833,
+      "learning_rate": 2e-05,
+      "loss": 0.1054,
+      "step": 148
+    },
+    {
+      "epoch": 1.2714666666666667,
+      "grad_norm": 0.04423248714119304,
+      "learning_rate": 2e-05,
+      "loss": 0.0841,
+      "step": 149
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.049931966607835034,
+      "learning_rate": 2e-05,
+      "loss": 0.0961,
+      "step": 150
+    },
+    {
+      "epoch": 1.2885333333333333,
+      "grad_norm": 0.06178656953298769,
+      "learning_rate": 2e-05,
+      "loss": 0.0854,
+      "step": 151
+    },
+    {
+      "epoch": 1.2970666666666666,
+      "grad_norm": 0.05783812343287897,
+      "learning_rate": 2e-05,
+      "loss": 0.1141,
+      "step": 152
+    },
+    {
+      "epoch": 1.3056,
+      "grad_norm": 0.048955120400167584,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 153
+    },
+    {
+      "epoch": 1.3141333333333334,
+      "grad_norm": 0.12270174746806978,
+      "learning_rate": 2e-05,
+      "loss": 0.1553,
+      "step": 154
+    },
+    {
+      "epoch": 1.3226666666666667,
+      "grad_norm": 0.06928026959973474,
+      "learning_rate": 2e-05,
+      "loss": 0.1274,
+      "step": 155
+    },
+    {
+      "epoch": 1.3312,
+      "grad_norm": 0.04756100666105405,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 156
+    },
+    {
+      "epoch": 1.3397333333333332,
+      "grad_norm": 0.056054951338196934,
+      "learning_rate": 2e-05,
+      "loss": 0.0831,
+      "step": 157
+    },
+    {
+      "epoch": 1.3482666666666667,
+      "grad_norm": 0.0516990471964239,
+      "learning_rate": 2e-05,
+      "loss": 0.0883,
+      "step": 158
+    },
+    {
+      "epoch": 1.3568,
+      "grad_norm": 0.06011650542069954,
+      "learning_rate": 2e-05,
+      "loss": 0.0938,
+      "step": 159
+    },
+    {
+      "epoch": 1.3653333333333333,
+      "grad_norm": 0.051831307951873976,
+      "learning_rate": 2e-05,
+      "loss": 0.1019,
+      "step": 160
+    },
+    {
+      "epoch": 1.3738666666666668,
+      "grad_norm": 0.0544902702048936,
+      "learning_rate": 2e-05,
+      "loss": 0.0906,
+      "step": 161
+    },
+    {
+      "epoch": 1.3824,
+      "grad_norm": 0.06057617032526,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 162
+    },
+    {
+      "epoch": 1.3909333333333334,
+      "grad_norm": 0.07288058025189605,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 163
+    },
+    {
+      "epoch": 1.3994666666666666,
+      "grad_norm": 0.05639043792084219,
+      "learning_rate": 2e-05,
+      "loss": 0.1031,
+      "step": 164
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.0586469408837505,
+      "learning_rate": 2e-05,
+      "loss": 0.1143,
+      "step": 165
+    },
+    {
+      "epoch": 1.4165333333333332,
+      "grad_norm": 0.059243429040783904,
+      "learning_rate": 2e-05,
+      "loss": 0.0838,
+      "step": 166
+    },
+    {
+      "epoch": 1.4250666666666667,
+      "grad_norm": 0.0621476448363388,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 167
+    },
+    {
+      "epoch": 1.4336,
+      "grad_norm": 0.06083867682720169,
+      "learning_rate": 2e-05,
+      "loss": 0.1119,
+      "step": 168
+    },
+    {
+      "epoch": 1.4421333333333333,
+      "grad_norm": 0.09583165335305677,
+      "learning_rate": 2e-05,
+      "loss": 0.1028,
+      "step": 169
+    },
+    {
+      "epoch": 1.4506666666666668,
+      "grad_norm": 0.06411638581314043,
+      "learning_rate": 2e-05,
+      "loss": 0.1181,
+      "step": 170
+    },
+    {
+      "epoch": 1.4592,
+      "grad_norm": 0.05632977234908945,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 171
+    },
+    {
+      "epoch": 1.4677333333333333,
+      "grad_norm": 0.05666068971337413,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 172
+    },
+    {
+      "epoch": 1.4762666666666666,
+      "grad_norm": 0.04502062723807536,
+      "learning_rate": 2e-05,
+      "loss": 0.0588,
+      "step": 173
+    },
+    {
+      "epoch": 1.4848,
+      "grad_norm": 0.05916500176868301,
+      "learning_rate": 2e-05,
+      "loss": 0.0949,
+      "step": 174
+    },
+    {
+      "epoch": 1.4933333333333334,
+      "grad_norm": 0.056484273808864845,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 175
+    },
+    {
+      "epoch": 1.5018666666666667,
+      "grad_norm": 0.06652084448571674,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 176
+    },
+    {
+      "epoch": 1.5104,
+      "grad_norm": 0.05400238078068516,
+      "learning_rate": 2e-05,
+      "loss": 0.0919,
+      "step": 177
+    },
+    {
+      "epoch": 1.5189333333333335,
+      "grad_norm": 0.04979579743346662,
+      "learning_rate": 2e-05,
+      "loss": 0.0879,
+      "step": 178
+    },
+    {
+      "epoch": 1.5274666666666668,
+      "grad_norm": 0.06876105414733971,
+      "learning_rate": 2e-05,
+      "loss": 0.1162,
+      "step": 179
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.05633262015499721,
+      "learning_rate": 2e-05,
+      "loss": 0.1142,
+      "step": 180
+    },
+    {
+      "epoch": 1.5445333333333333,
+      "grad_norm": 0.0599508967519892,
+      "learning_rate": 2e-05,
+      "loss": 0.1073,
+      "step": 181
+    },
+    {
+      "epoch": 1.5530666666666666,
+      "grad_norm": 0.058912170976454126,
+      "learning_rate": 2e-05,
+      "loss": 0.1102,
+      "step": 182
+    },
+    {
+      "epoch": 1.5615999999999999,
+      "grad_norm": 0.05361414988566254,
+      "learning_rate": 2e-05,
+      "loss": 0.0885,
+      "step": 183
+    },
+    {
+      "epoch": 1.5701333333333334,
+      "grad_norm": 0.04517277847384077,
+      "learning_rate": 2e-05,
+      "loss": 0.0763,
+      "step": 184
+    },
+    {
+      "epoch": 1.5786666666666667,
+      "grad_norm": 0.05551553662051771,
+      "learning_rate": 2e-05,
+      "loss": 0.0877,
+      "step": 185
+    },
+    {
+      "epoch": 1.5872000000000002,
+      "grad_norm": 0.05814223969236194,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 186
+    },
+    {
+      "epoch": 1.5957333333333334,
+      "grad_norm": 0.05710054166191597,
+      "learning_rate": 2e-05,
+      "loss": 0.0962,
+      "step": 187
+    },
+    {
+      "epoch": 1.6042666666666667,
+      "grad_norm": 0.054744343932104075,
+      "learning_rate": 2e-05,
+      "loss": 0.0873,
+      "step": 188
+    },
+    {
+      "epoch": 1.6128,
+      "grad_norm": 0.051145521687090995,
+      "learning_rate": 2e-05,
+      "loss": 0.0855,
+      "step": 189
+    },
+    {
+      "epoch": 1.6213333333333333,
+      "grad_norm": 0.05414658860737789,
+      "learning_rate": 2e-05,
+      "loss": 0.0872,
+      "step": 190
+    },
+    {
+      "epoch": 1.6298666666666666,
+      "grad_norm": 0.05084744595533029,
+      "learning_rate": 2e-05,
+      "loss": 0.0891,
+      "step": 191
+    },
+    {
+      "epoch": 1.6383999999999999,
+      "grad_norm": 0.0567070082820597,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 192
+    },
+    {
+      "epoch": 1.6469333333333334,
+      "grad_norm": 0.0494785311411315,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 193
+    },
+    {
+      "epoch": 1.6554666666666666,
+      "grad_norm": 0.062341158530385396,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 194
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.059888336716275685,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 195
+    },
+    {
+      "epoch": 1.6725333333333334,
+      "grad_norm": 0.07346562318829057,
+      "learning_rate": 2e-05,
+      "loss": 0.1329,
+      "step": 196
+    },
+    {
+      "epoch": 1.6810666666666667,
+      "grad_norm": 0.0792360016934733,
+      "learning_rate": 2e-05,
+      "loss": 0.1392,
+      "step": 197
+    },
+    {
+      "epoch": 1.6896,
+      "grad_norm": 0.0800342963229883,
+      "learning_rate": 2e-05,
+      "loss": 0.1199,
+      "step": 198
+    },
+    {
+      "epoch": 1.6981333333333333,
+      "grad_norm": 0.06848045933195548,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 199
+    },
+    {
+      "epoch": 1.7066666666666666,
+      "grad_norm": 0.05743199019316764,
+      "learning_rate": 2e-05,
+      "loss": 0.0811,
+      "step": 200
+    },
+    {
+      "epoch": 1.7151999999999998,
+      "grad_norm": 0.07170531168284446,
+      "learning_rate": 2e-05,
+      "loss": 0.1079,
+      "step": 201
+    },
+    {
+      "epoch": 1.7237333333333333,
+      "grad_norm": 0.05772905481368506,
+      "learning_rate": 2e-05,
+      "loss": 0.0844,
+      "step": 202
+    },
+    {
+      "epoch": 1.7322666666666666,
+      "grad_norm": 0.07504946014098464,
+      "learning_rate": 2e-05,
+      "loss": 0.1257,
+      "step": 203
+    },
+    {
+      "epoch": 1.7408000000000001,
+      "grad_norm": 0.06450179819785838,
+      "learning_rate": 2e-05,
+      "loss": 0.1104,
+      "step": 204
+    },
+    {
+      "epoch": 1.7493333333333334,
+      "grad_norm": 0.06595445965110332,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 205
+    },
+    {
+      "epoch": 1.7578666666666667,
+      "grad_norm": 0.07203558121131749,
+      "learning_rate": 2e-05,
+      "loss": 0.1117,
+      "step": 206
+    },
+    {
+      "epoch": 1.7664,
+      "grad_norm": 0.05954646782409283,
+      "learning_rate": 2e-05,
+      "loss": 0.0729,
+      "step": 207
+    },
+    {
+      "epoch": 1.7749333333333333,
+      "grad_norm": 0.06624894584410884,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 208
+    },
+    {
+      "epoch": 1.7834666666666665,
+      "grad_norm": 0.06888562028256219,
+      "learning_rate": 2e-05,
+      "loss": 0.1398,
+      "step": 209
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.061224088077794406,
+      "learning_rate": 2e-05,
+      "loss": 0.1112,
+      "step": 210
+    },
+    {
+      "epoch": 1.8005333333333333,
+      "grad_norm": 0.06857358883856608,
+      "learning_rate": 2e-05,
+      "loss": 0.1293,
+      "step": 211
+    },
+    {
+      "epoch": 1.8090666666666668,
+      "grad_norm": 0.06177352416779139,
+      "learning_rate": 2e-05,
+      "loss": 0.0884,
+      "step": 212
+    },
+    {
+      "epoch": 1.8176,
+      "grad_norm": 0.08324567429925228,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 213
+    },
+    {
+      "epoch": 1.8261333333333334,
+      "grad_norm": 0.06771677297787752,
+      "learning_rate": 2e-05,
+      "loss": 0.089,
+      "step": 214
+    },
+    {
+      "epoch": 1.8346666666666667,
+      "grad_norm": 0.07055754809472485,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 215
+    },
+    {
+      "epoch": 1.8432,
+      "grad_norm": 0.05856797724392531,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 216
+    },
+    {
+      "epoch": 1.8517333333333332,
+      "grad_norm": 0.07555286129801597,
+      "learning_rate": 2e-05,
+      "loss": 0.0913,
+      "step": 217
+    },
+    {
+      "epoch": 1.8602666666666665,
+      "grad_norm": 0.09242462538643775,
+      "learning_rate": 2e-05,
+      "loss": 0.1241,
+      "step": 218
+    },
+    {
+      "epoch": 1.8688,
+      "grad_norm": 0.06550805537088374,
+      "learning_rate": 2e-05,
+      "loss": 0.0819,
+      "step": 219
+    },
+    {
+      "epoch": 1.8773333333333333,
+      "grad_norm": 0.06016048263236861,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 220
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 351,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.1374063827523994e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-220/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa4edad1da5557fcd52a6da980443588016cead6f0444a3562cfa68029c66a04
+size 6840

checkpoint-220/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

checkpoint-240/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ../ckpts/Meta-Llama-3-8B-Instruct
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.11.1

checkpoint-240/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "../ckpts/Meta-Llama-3-8B-Instruct",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "gate_proj",
+    "q_proj",
+    "lm_head",
+    "o_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-240/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d2b0102ec16154b1882251190f7984562a5f9255aaed7799bf80b27e06299967
+size 1138856856

checkpoint-240/global_step240/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2d2d9813bc8f9fdb7e61aa7c515b974f9e3a321c0c00c461ae2223c64ea368ac
+size 528781328

checkpoint-240/global_step240/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b92dc14f0a23e1d44580987f2a390d1ffcd884ae2bbee1358798e9f91fbece65
+size 199905337

checkpoint-240/latest

@@ -0,0 +1 @@
+global_step240

checkpoint-240/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:520b20ae855dee764163de1cce297ae24a311e9fc58dab9dc56d3069fcb4f3dc
+size 14244

checkpoint-240/trainer_state.json

@@ -0,0 +1,1713 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.048,
+  "eval_steps": 500,
+  "global_step": 240,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008533333333333334,
+      "grad_norm": 100.21848203113535,
+      "learning_rate": 0.0,
+      "loss": 7.1962,
+      "step": 1
+    },
+    {
+      "epoch": 0.017066666666666667,
+      "grad_norm": 96.43006188910957,
+      "learning_rate": 3.0102999566398115e-06,
+      "loss": 6.9414,
+      "step": 2
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 97.35803466618715,
+      "learning_rate": 4.771212547196624e-06,
+      "loss": 7.0102,
+      "step": 3
+    },
+    {
+      "epoch": 0.034133333333333335,
+      "grad_norm": 95.14837816372646,
+      "learning_rate": 6.020599913279623e-06,
+      "loss": 6.5295,
+      "step": 4
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 91.76544275692784,
+      "learning_rate": 6.989700043360187e-06,
+      "loss": 6.4806,
+      "step": 5
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 84.4494318688335,
+      "learning_rate": 7.781512503836437e-06,
+      "loss": 6.4194,
+      "step": 6
+    },
+    {
+      "epoch": 0.05973333333333333,
+      "grad_norm": 71.37977490595638,
+      "learning_rate": 8.450980400142568e-06,
+      "loss": 5.4953,
+      "step": 7
+    },
+    {
+      "epoch": 0.06826666666666667,
+      "grad_norm": 49.31153456754566,
+      "learning_rate": 9.030899869919434e-06,
+      "loss": 5.4123,
+      "step": 8
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 20.37296364560341,
+      "learning_rate": 9.542425094393249e-06,
+      "loss": 5.2334,
+      "step": 9
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 7.968467079076881,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 5.0282,
+      "step": 10
+    },
+    {
+      "epoch": 0.09386666666666667,
+      "grad_norm": 3.559446532055649,
+      "learning_rate": 1.041392685158225e-05,
+      "loss": 4.612,
+      "step": 11
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 3.5528846947995674,
+      "learning_rate": 1.0791812460476248e-05,
+      "loss": 4.9475,
+      "step": 12
+    },
+    {
+      "epoch": 0.11093333333333333,
+      "grad_norm": 3.541968897471334,
+      "learning_rate": 1.1139433523068365e-05,
+      "loss": 4.2777,
+      "step": 13
+    },
+    {
+      "epoch": 0.11946666666666667,
+      "grad_norm": 3.54718070036198,
+      "learning_rate": 1.1461280356782378e-05,
+      "loss": 4.3507,
+      "step": 14
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 3.8632334830606747,
+      "learning_rate": 1.1760912590556813e-05,
+      "loss": 4.5364,
+      "step": 15
+    },
+    {
+      "epoch": 0.13653333333333334,
+      "grad_norm": 3.6637424744054004,
+      "learning_rate": 1.2041199826559246e-05,
+      "loss": 3.9672,
+      "step": 16
+    },
+    {
+      "epoch": 0.14506666666666668,
+      "grad_norm": 3.919802904818311,
+      "learning_rate": 1.230448921378274e-05,
+      "loss": 4.0618,
+      "step": 17
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 4.71904950738746,
+      "learning_rate": 1.2552725051033058e-05,
+      "loss": 4.6656,
+      "step": 18
+    },
+    {
+      "epoch": 0.16213333333333332,
+      "grad_norm": 4.6656317698690835,
+      "learning_rate": 1.2787536009528288e-05,
+      "loss": 4.1131,
+      "step": 19
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 5.145138692367417,
+      "learning_rate": 1.301029995663981e-05,
+      "loss": 4.0989,
+      "step": 20
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 5.923538014759818,
+      "learning_rate": 1.3222192947339192e-05,
+      "loss": 4.4991,
+      "step": 21
+    },
+    {
+      "epoch": 0.18773333333333334,
+      "grad_norm": 5.941056962941364,
+      "learning_rate": 1.3424226808222062e-05,
+      "loss": 4.0836,
+      "step": 22
+    },
+    {
+      "epoch": 0.19626666666666667,
+      "grad_norm": 6.171026012117947,
+      "learning_rate": 1.3617278360175927e-05,
+      "loss": 3.6861,
+      "step": 23
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 7.130542138930838,
+      "learning_rate": 1.380211241711606e-05,
+      "loss": 4.0958,
+      "step": 24
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 7.328837606110418,
+      "learning_rate": 1.3979400086720374e-05,
+      "loss": 3.9524,
+      "step": 25
+    },
+    {
+      "epoch": 0.22186666666666666,
+      "grad_norm": 6.923489005711429,
+      "learning_rate": 1.4149733479708178e-05,
+      "loss": 3.6062,
+      "step": 26
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 6.409498926059221,
+      "learning_rate": 1.4313637641589872e-05,
+      "loss": 3.2034,
+      "step": 27
+    },
+    {
+      "epoch": 0.23893333333333333,
+      "grad_norm": 5.781628405584682,
+      "learning_rate": 1.4471580313422191e-05,
+      "loss": 2.8158,
+      "step": 28
+    },
+    {
+      "epoch": 0.24746666666666667,
+      "grad_norm": 6.2927590068243315,
+      "learning_rate": 1.4623979978989559e-05,
+      "loss": 2.9803,
+      "step": 29
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 6.103844678630006,
+      "learning_rate": 1.4771212547196623e-05,
+      "loss": 2.847,
+      "step": 30
+    },
+    {
+      "epoch": 0.26453333333333334,
+      "grad_norm": 7.656341925867144,
+      "learning_rate": 1.4913616938342726e-05,
+      "loss": 3.0907,
+      "step": 31
+    },
+    {
+      "epoch": 0.2730666666666667,
+      "grad_norm": 6.324242877501844,
+      "learning_rate": 1.5051499783199059e-05,
+      "loss": 2.3467,
+      "step": 32
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 7.606313101162739,
+      "learning_rate": 1.5185139398778874e-05,
+      "loss": 2.5292,
+      "step": 33
+    },
+    {
+      "epoch": 0.29013333333333335,
+      "grad_norm": 8.553792493849265,
+      "learning_rate": 1.531478917042255e-05,
+      "loss": 2.4547,
+      "step": 34
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 8.483368703272543,
+      "learning_rate": 1.5440680443502753e-05,
+      "loss": 2.1956,
+      "step": 35
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 8.364739352838077,
+      "learning_rate": 1.5563025007672873e-05,
+      "loss": 1.8552,
+      "step": 36
+    },
+    {
+      "epoch": 0.3157333333333333,
+      "grad_norm": 9.37663682000104,
+      "learning_rate": 1.5682017240669948e-05,
+      "loss": 1.9228,
+      "step": 37
+    },
+    {
+      "epoch": 0.32426666666666665,
+      "grad_norm": 9.823047193440066,
+      "learning_rate": 1.57978359661681e-05,
+      "loss": 1.7033,
+      "step": 38
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 9.692618512955894,
+      "learning_rate": 1.591064607026499e-05,
+      "loss": 1.3768,
+      "step": 39
+    },
+    {
+      "epoch": 0.3413333333333333,
+      "grad_norm": 9.08889021911031,
+      "learning_rate": 1.6020599913279622e-05,
+      "loss": 1.3015,
+      "step": 40
+    },
+    {
+      "epoch": 0.34986666666666666,
+      "grad_norm": 8.081534221516058,
+      "learning_rate": 1.6127838567197353e-05,
+      "loss": 0.9228,
+      "step": 41
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 6.238638048950311,
+      "learning_rate": 1.6232492903979005e-05,
+      "loss": 0.7267,
+      "step": 42
+    },
+    {
+      "epoch": 0.36693333333333333,
+      "grad_norm": 3.4058036861773604,
+      "learning_rate": 1.6334684555795865e-05,
+      "loss": 0.5875,
+      "step": 43
+    },
+    {
+      "epoch": 0.37546666666666667,
+      "grad_norm": 2.079163829467713,
+      "learning_rate": 1.6434526764861872e-05,
+      "loss": 0.6355,
+      "step": 44
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 1.5597487833024746,
+      "learning_rate": 1.6532125137753435e-05,
+      "loss": 0.5106,
+      "step": 45
+    },
+    {
+      "epoch": 0.39253333333333335,
+      "grad_norm": 2.491689602375256,
+      "learning_rate": 1.662757831681574e-05,
+      "loss": 0.6454,
+      "step": 46
+    },
+    {
+      "epoch": 0.4010666666666667,
+      "grad_norm": 2.010880438195854,
+      "learning_rate": 1.672097857935717e-05,
+      "loss": 0.4757,
+      "step": 47
+    },
+    {
+      "epoch": 0.4096,
+      "grad_norm": 1.9452805114322096,
+      "learning_rate": 1.681241237375587e-05,
+      "loss": 0.4133,
+      "step": 48
+    },
+    {
+      "epoch": 0.41813333333333336,
+      "grad_norm": 1.7620848552306103,
+      "learning_rate": 1.6901960800285137e-05,
+      "loss": 0.4004,
+      "step": 49
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 1.278224489774809,
+      "learning_rate": 1.6989700043360187e-05,
+      "loss": 0.3523,
+      "step": 50
+    },
+    {
+      "epoch": 0.4352,
+      "grad_norm": 1.6151354758303231,
+      "learning_rate": 1.7075701760979363e-05,
+      "loss": 0.4317,
+      "step": 51
+    },
+    {
+      "epoch": 0.4437333333333333,
+      "grad_norm": 1.3451396055695035,
+      "learning_rate": 1.716003343634799e-05,
+      "loss": 0.3474,
+      "step": 52
+    },
+    {
+      "epoch": 0.45226666666666665,
+      "grad_norm": 1.6814977782362666,
+      "learning_rate": 1.724275869600789e-05,
+      "loss": 0.3706,
+      "step": 53
+    },
+    {
+      "epoch": 0.4608,
+      "grad_norm": 1.1682442432688667,
+      "learning_rate": 1.7323937598229687e-05,
+      "loss": 0.3488,
+      "step": 54
+    },
+    {
+      "epoch": 0.4693333333333333,
+      "grad_norm": 0.8839814540462471,
+      "learning_rate": 1.7403626894942437e-05,
+      "loss": 0.293,
+      "step": 55
+    },
+    {
+      "epoch": 0.47786666666666666,
+      "grad_norm": 0.7974304806295485,
+      "learning_rate": 1.7481880270062003e-05,
+      "loss": 0.2717,
+      "step": 56
+    },
+    {
+      "epoch": 0.4864,
+      "grad_norm": 0.6232203657585239,
+      "learning_rate": 1.7558748556724913e-05,
+      "loss": 0.1741,
+      "step": 57
+    },
+    {
+      "epoch": 0.49493333333333334,
+      "grad_norm": 0.6850688604563008,
+      "learning_rate": 1.763427993562937e-05,
+      "loss": 0.228,
+      "step": 58
+    },
+    {
+      "epoch": 0.5034666666666666,
+      "grad_norm": 0.5923826384300431,
+      "learning_rate": 1.7708520116421443e-05,
+      "loss": 0.2131,
+      "step": 59
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.489754430485032,
+      "learning_rate": 1.7781512503836432e-05,
+      "loss": 0.165,
+      "step": 60
+    },
+    {
+      "epoch": 0.5205333333333333,
+      "grad_norm": 0.5280940052395061,
+      "learning_rate": 1.7853298350107667e-05,
+      "loss": 0.1658,
+      "step": 61
+    },
+    {
+      "epoch": 0.5290666666666667,
+      "grad_norm": 0.4750905992036739,
+      "learning_rate": 1.7923916894982537e-05,
+      "loss": 0.1438,
+      "step": 62
+    },
+    {
+      "epoch": 0.5376,
+      "grad_norm": 0.4162998392722401,
+      "learning_rate": 1.7993405494535815e-05,
+      "loss": 0.1555,
+      "step": 63
+    },
+    {
+      "epoch": 0.5461333333333334,
+      "grad_norm": 0.26863266246370443,
+      "learning_rate": 1.806179973983887e-05,
+      "loss": 0.1323,
+      "step": 64
+    },
+    {
+      "epoch": 0.5546666666666666,
+      "grad_norm": 0.26534221125601215,
+      "learning_rate": 1.8129133566428553e-05,
+      "loss": 0.1671,
+      "step": 65
+    },
+    {
+      "epoch": 0.5632,
+      "grad_norm": 0.2548662962257576,
+      "learning_rate": 1.8195439355418686e-05,
+      "loss": 0.1308,
+      "step": 66
+    },
+    {
+      "epoch": 0.5717333333333333,
+      "grad_norm": 0.18045394638578796,
+      "learning_rate": 1.8260748027008263e-05,
+      "loss": 0.1262,
+      "step": 67
+    },
+    {
+      "epoch": 0.5802666666666667,
+      "grad_norm": 0.17070388073612064,
+      "learning_rate": 1.8325089127062364e-05,
+      "loss": 0.1192,
+      "step": 68
+    },
+    {
+      "epoch": 0.5888,
+      "grad_norm": 0.1531381679244776,
+      "learning_rate": 1.8388490907372553e-05,
+      "loss": 0.1274,
+      "step": 69
+    },
+    {
+      "epoch": 0.5973333333333334,
+      "grad_norm": 0.18196434993687946,
+      "learning_rate": 1.8450980400142568e-05,
+      "loss": 0.1375,
+      "step": 70
+    },
+    {
+      "epoch": 0.6058666666666667,
+      "grad_norm": 0.15324416972951205,
+      "learning_rate": 1.8512583487190752e-05,
+      "loss": 0.1599,
+      "step": 71
+    },
+    {
+      "epoch": 0.6144,
+      "grad_norm": 0.10884462064503801,
+      "learning_rate": 1.857332496431268e-05,
+      "loss": 0.1041,
+      "step": 72
+    },
+    {
+      "epoch": 0.6229333333333333,
+      "grad_norm": 0.12915133528192668,
+      "learning_rate": 1.8633228601204555e-05,
+      "loss": 0.1406,
+      "step": 73
+    },
+    {
+      "epoch": 0.6314666666666666,
+      "grad_norm": 0.12553425699952878,
+      "learning_rate": 1.8692317197309763e-05,
+      "loss": 0.1256,
+      "step": 74
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11976960918968543,
+      "learning_rate": 1.8750612633916997e-05,
+      "loss": 0.1144,
+      "step": 75
+    },
+    {
+      "epoch": 0.6485333333333333,
+      "grad_norm": 0.115805998298789,
+      "learning_rate": 1.8808135922807914e-05,
+      "loss": 0.1528,
+      "step": 76
+    },
+    {
+      "epoch": 0.6570666666666667,
+      "grad_norm": 0.10325948496697443,
+      "learning_rate": 1.8864907251724818e-05,
+      "loss": 0.1044,
+      "step": 77
+    },
+    {
+      "epoch": 0.6656,
+      "grad_norm": 0.09595064346541006,
+      "learning_rate": 1.8920946026904802e-05,
+      "loss": 0.1534,
+      "step": 78
+    },
+    {
+      "epoch": 0.6741333333333334,
+      "grad_norm": 0.08796742845240496,
+      "learning_rate": 1.8976270912904414e-05,
+      "loss": 0.1155,
+      "step": 79
+    },
+    {
+      "epoch": 0.6826666666666666,
+      "grad_norm": 0.08218991738379527,
+      "learning_rate": 1.9030899869919434e-05,
+      "loss": 0.1311,
+      "step": 80
+    },
+    {
+      "epoch": 0.6912,
+      "grad_norm": 0.08290815261109215,
+      "learning_rate": 1.9084850188786497e-05,
+      "loss": 0.11,
+      "step": 81
+    },
+    {
+      "epoch": 0.6997333333333333,
+      "grad_norm": 0.0794132180176064,
+      "learning_rate": 1.9138138523837165e-05,
+      "loss": 0.1135,
+      "step": 82
+    },
+    {
+      "epoch": 0.7082666666666667,
+      "grad_norm": 0.06934410705255296,
+      "learning_rate": 1.919078092376074e-05,
+      "loss": 0.109,
+      "step": 83
+    },
+    {
+      "epoch": 0.7168,
+      "grad_norm": 0.09000563031870593,
+      "learning_rate": 1.9242792860618813e-05,
+      "loss": 0.12,
+      "step": 84
+    },
+    {
+      "epoch": 0.7253333333333334,
+      "grad_norm": 0.1134042277879818,
+      "learning_rate": 1.929418925714293e-05,
+      "loss": 0.1223,
+      "step": 85
+    },
+    {
+      "epoch": 0.7338666666666667,
+      "grad_norm": 0.09118764690233076,
+      "learning_rate": 1.9344984512435673e-05,
+      "loss": 0.1459,
+      "step": 86
+    },
+    {
+      "epoch": 0.7424,
+      "grad_norm": 0.07873016754353963,
+      "learning_rate": 1.9395192526186183e-05,
+      "loss": 0.1422,
+      "step": 87
+    },
+    {
+      "epoch": 0.7509333333333333,
+      "grad_norm": 0.1796495874463076,
+      "learning_rate": 1.9444826721501687e-05,
+      "loss": 0.1291,
+      "step": 88
+    },
+    {
+      "epoch": 0.7594666666666666,
+      "grad_norm": 0.0679589944174269,
+      "learning_rate": 1.9493900066449125e-05,
+      "loss": 0.108,
+      "step": 89
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.08174688574235538,
+      "learning_rate": 1.9542425094393246e-05,
+      "loss": 0.1081,
+      "step": 90
+    },
+    {
+      "epoch": 0.7765333333333333,
+      "grad_norm": 0.057137370501406756,
+      "learning_rate": 1.9590413923210934e-05,
+      "loss": 0.0934,
+      "step": 91
+    },
+    {
+      "epoch": 0.7850666666666667,
+      "grad_norm": 0.06578111924908255,
+      "learning_rate": 1.9637878273455555e-05,
+      "loss": 0.1085,
+      "step": 92
+    },
+    {
+      "epoch": 0.7936,
+      "grad_norm": 0.08945990540906254,
+      "learning_rate": 1.968482948553935e-05,
+      "loss": 0.1747,
+      "step": 93
+    },
+    {
+      "epoch": 0.8021333333333334,
+      "grad_norm": 0.06183863311044229,
+      "learning_rate": 1.9731278535996986e-05,
+      "loss": 0.1136,
+      "step": 94
+    },
+    {
+      "epoch": 0.8106666666666666,
+      "grad_norm": 0.05777899602544702,
+      "learning_rate": 1.9777236052888476e-05,
+      "loss": 0.0984,
+      "step": 95
+    },
+    {
+      "epoch": 0.8192,
+      "grad_norm": 0.08130851607693534,
+      "learning_rate": 1.9822712330395683e-05,
+      "loss": 0.187,
+      "step": 96
+    },
+    {
+      "epoch": 0.8277333333333333,
+      "grad_norm": 0.06426546202002927,
+      "learning_rate": 1.986771734266245e-05,
+      "loss": 0.1296,
+      "step": 97
+    },
+    {
+      "epoch": 0.8362666666666667,
+      "grad_norm": 0.069692313707994,
+      "learning_rate": 1.991226075692495e-05,
+      "loss": 0.1404,
+      "step": 98
+    },
+    {
+      "epoch": 0.8448,
+      "grad_norm": 0.05494542266886729,
+      "learning_rate": 1.9956351945975496e-05,
+      "loss": 0.116,
+      "step": 99
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.07571686966840627,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 0.1539,
+      "step": 100
+    },
+    {
+      "epoch": 0.8618666666666667,
+      "grad_norm": 0.054351059117603705,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 101
+    },
+    {
+      "epoch": 0.8704,
+      "grad_norm": 0.06531899611551092,
+      "learning_rate": 2e-05,
+      "loss": 0.0827,
+      "step": 102
+    },
+    {
+      "epoch": 0.8789333333333333,
+      "grad_norm": 0.06131678646504652,
+      "learning_rate": 2e-05,
+      "loss": 0.1266,
+      "step": 103
+    },
+    {
+      "epoch": 0.8874666666666666,
+      "grad_norm": 0.06850220540661824,
+      "learning_rate": 2e-05,
+      "loss": 0.1456,
+      "step": 104
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.05806908951252483,
+      "learning_rate": 2e-05,
+      "loss": 0.0954,
+      "step": 105
+    },
+    {
+      "epoch": 0.9045333333333333,
+      "grad_norm": 0.06503642452033717,
+      "learning_rate": 2e-05,
+      "loss": 0.1417,
+      "step": 106
+    },
+    {
+      "epoch": 0.9130666666666667,
+      "grad_norm": 0.050486271853277066,
+      "learning_rate": 2e-05,
+      "loss": 0.0959,
+      "step": 107
+    },
+    {
+      "epoch": 0.9216,
+      "grad_norm": 0.07746063813802379,
+      "learning_rate": 2e-05,
+      "loss": 0.1256,
+      "step": 108
+    },
+    {
+      "epoch": 0.9301333333333334,
+      "grad_norm": 0.051231172380840004,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 109
+    },
+    {
+      "epoch": 0.9386666666666666,
+      "grad_norm": 0.056296443557859455,
+      "learning_rate": 2e-05,
+      "loss": 0.1056,
+      "step": 110
+    },
+    {
+      "epoch": 0.9472,
+      "grad_norm": 0.05058663240713958,
+      "learning_rate": 2e-05,
+      "loss": 0.0971,
+      "step": 111
+    },
+    {
+      "epoch": 0.9557333333333333,
+      "grad_norm": 0.05532886570130611,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 112
+    },
+    {
+      "epoch": 0.9642666666666667,
+      "grad_norm": 0.05327811326654907,
+      "learning_rate": 2e-05,
+      "loss": 0.0989,
+      "step": 113
+    },
+    {
+      "epoch": 0.9728,
+      "grad_norm": 0.05663279364147864,
+      "learning_rate": 2e-05,
+      "loss": 0.0958,
+      "step": 114
+    },
+    {
+      "epoch": 0.9813333333333333,
+      "grad_norm": 0.04930904541225805,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 115
+    },
+    {
+      "epoch": 0.9898666666666667,
+      "grad_norm": 0.06096947951115022,
+      "learning_rate": 2e-05,
+      "loss": 0.106,
+      "step": 116
+    },
+    {
+      "epoch": 0.9984,
+      "grad_norm": 0.050092322361182495,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 117
+    },
+    {
+      "epoch": 1.0069333333333332,
+      "grad_norm": 0.04980408443758999,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 118
+    },
+    {
+      "epoch": 1.0154666666666667,
+      "grad_norm": 0.051183082721834305,
+      "learning_rate": 2e-05,
+      "loss": 0.1049,
+      "step": 119
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.04332220265802814,
+      "learning_rate": 2e-05,
+      "loss": 0.0983,
+      "step": 120
+    },
+    {
+      "epoch": 1.0325333333333333,
+      "grad_norm": 0.07211731499677299,
+      "learning_rate": 2e-05,
+      "loss": 0.1386,
+      "step": 121
+    },
+    {
+      "epoch": 1.0410666666666666,
+      "grad_norm": 0.06550870223740553,
+      "learning_rate": 2e-05,
+      "loss": 0.1334,
+      "step": 122
+    },
+    {
+      "epoch": 1.0496,
+      "grad_norm": 0.05331839690767287,
+      "learning_rate": 2e-05,
+      "loss": 0.1014,
+      "step": 123
+    },
+    {
+      "epoch": 1.0581333333333334,
+      "grad_norm": 0.05227685628767905,
+      "learning_rate": 2e-05,
+      "loss": 0.1098,
+      "step": 124
+    },
+    {
+      "epoch": 1.0666666666666667,
+      "grad_norm": 0.07641693882491171,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 125
+    },
+    {
+      "epoch": 1.0752,
+      "grad_norm": 0.052835367770791786,
+      "learning_rate": 2e-05,
+      "loss": 0.1231,
+      "step": 126
+    },
+    {
+      "epoch": 1.0837333333333334,
+      "grad_norm": 0.07520329755025788,
+      "learning_rate": 2e-05,
+      "loss": 0.085,
+      "step": 127
+    },
+    {
+      "epoch": 1.0922666666666667,
+      "grad_norm": 0.07670066152157425,
+      "learning_rate": 2e-05,
+      "loss": 0.1071,
+      "step": 128
+    },
+    {
+      "epoch": 1.1008,
+      "grad_norm": 0.052832906560645154,
+      "learning_rate": 2e-05,
+      "loss": 0.1093,
+      "step": 129
+    },
+    {
+      "epoch": 1.1093333333333333,
+      "grad_norm": 0.06573889037311398,
+      "learning_rate": 2e-05,
+      "loss": 0.1193,
+      "step": 130
+    },
+    {
+      "epoch": 1.1178666666666666,
+      "grad_norm": 0.05175471296566334,
+      "learning_rate": 2e-05,
+      "loss": 0.1184,
+      "step": 131
+    },
+    {
+      "epoch": 1.1264,
+      "grad_norm": 0.05912231419793496,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 132
+    },
+    {
+      "epoch": 1.1349333333333333,
+      "grad_norm": 0.04899140475981105,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 133
+    },
+    {
+      "epoch": 1.1434666666666666,
+      "grad_norm": 0.05939916939142137,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 134
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.0516819217599706,
+      "learning_rate": 2e-05,
+      "loss": 0.0834,
+      "step": 135
+    },
+    {
+      "epoch": 1.1605333333333334,
+      "grad_norm": 0.05456440346454737,
+      "learning_rate": 2e-05,
+      "loss": 0.1183,
+      "step": 136
+    },
+    {
+      "epoch": 1.1690666666666667,
+      "grad_norm": 0.059906464476343235,
+      "learning_rate": 2e-05,
+      "loss": 0.1048,
+      "step": 137
+    },
+    {
+      "epoch": 1.1776,
+      "grad_norm": 0.0720112680204319,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 138
+    },
+    {
+      "epoch": 1.1861333333333333,
+      "grad_norm": 0.04940202805828527,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 139
+    },
+    {
+      "epoch": 1.1946666666666665,
+      "grad_norm": 0.060088609545130046,
+      "learning_rate": 2e-05,
+      "loss": 0.0952,
+      "step": 140
+    },
+    {
+      "epoch": 1.2032,
+      "grad_norm": 0.04694761423612446,
+      "learning_rate": 2e-05,
+      "loss": 0.0717,
+      "step": 141
+    },
+    {
+      "epoch": 1.2117333333333333,
+      "grad_norm": 0.05628581562512457,
+      "learning_rate": 2e-05,
+      "loss": 0.1062,
+      "step": 142
+    },
+    {
+      "epoch": 1.2202666666666666,
+      "grad_norm": 0.06876420990437652,
+      "learning_rate": 2e-05,
+      "loss": 0.1218,
+      "step": 143
+    },
+    {
+      "epoch": 1.2288000000000001,
+      "grad_norm": 0.058774700501610655,
+      "learning_rate": 2e-05,
+      "loss": 0.1125,
+      "step": 144
+    },
+    {
+      "epoch": 1.2373333333333334,
+      "grad_norm": 0.061855922064341186,
+      "learning_rate": 2e-05,
+      "loss": 0.1295,
+      "step": 145
+    },
+    {
+      "epoch": 1.2458666666666667,
+      "grad_norm": 0.0813047704730138,
+      "learning_rate": 2e-05,
+      "loss": 0.1165,
+      "step": 146
+    },
+    {
+      "epoch": 1.2544,
+      "grad_norm": 0.061374000305305446,
+      "learning_rate": 2e-05,
+      "loss": 0.1094,
+      "step": 147
+    },
+    {
+      "epoch": 1.2629333333333332,
+      "grad_norm": 0.055537169110833,
+      "learning_rate": 2e-05,
+      "loss": 0.1054,
+      "step": 148
+    },
+    {
+      "epoch": 1.2714666666666667,
+      "grad_norm": 0.04423248714119304,
+      "learning_rate": 2e-05,
+      "loss": 0.0841,
+      "step": 149
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.049931966607835034,
+      "learning_rate": 2e-05,
+      "loss": 0.0961,
+      "step": 150
+    },
+    {
+      "epoch": 1.2885333333333333,
+      "grad_norm": 0.06178656953298769,
+      "learning_rate": 2e-05,
+      "loss": 0.0854,
+      "step": 151
+    },
+    {
+      "epoch": 1.2970666666666666,
+      "grad_norm": 0.05783812343287897,
+      "learning_rate": 2e-05,
+      "loss": 0.1141,
+      "step": 152
+    },
+    {
+      "epoch": 1.3056,
+      "grad_norm": 0.048955120400167584,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 153
+    },
+    {
+      "epoch": 1.3141333333333334,
+      "grad_norm": 0.12270174746806978,
+      "learning_rate": 2e-05,
+      "loss": 0.1553,
+      "step": 154
+    },
+    {
+      "epoch": 1.3226666666666667,
+      "grad_norm": 0.06928026959973474,
+      "learning_rate": 2e-05,
+      "loss": 0.1274,
+      "step": 155
+    },
+    {
+      "epoch": 1.3312,
+      "grad_norm": 0.04756100666105405,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 156
+    },
+    {
+      "epoch": 1.3397333333333332,
+      "grad_norm": 0.056054951338196934,
+      "learning_rate": 2e-05,
+      "loss": 0.0831,
+      "step": 157
+    },
+    {
+      "epoch": 1.3482666666666667,
+      "grad_norm": 0.0516990471964239,
+      "learning_rate": 2e-05,
+      "loss": 0.0883,
+      "step": 158
+    },
+    {
+      "epoch": 1.3568,
+      "grad_norm": 0.06011650542069954,
+      "learning_rate": 2e-05,
+      "loss": 0.0938,
+      "step": 159
+    },
+    {
+      "epoch": 1.3653333333333333,
+      "grad_norm": 0.051831307951873976,
+      "learning_rate": 2e-05,
+      "loss": 0.1019,
+      "step": 160
+    },
+    {
+      "epoch": 1.3738666666666668,
+      "grad_norm": 0.0544902702048936,
+      "learning_rate": 2e-05,
+      "loss": 0.0906,
+      "step": 161
+    },
+    {
+      "epoch": 1.3824,
+      "grad_norm": 0.06057617032526,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 162
+    },
+    {
+      "epoch": 1.3909333333333334,
+      "grad_norm": 0.07288058025189605,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 163
+    },
+    {
+      "epoch": 1.3994666666666666,
+      "grad_norm": 0.05639043792084219,
+      "learning_rate": 2e-05,
+      "loss": 0.1031,
+      "step": 164
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.0586469408837505,
+      "learning_rate": 2e-05,
+      "loss": 0.1143,
+      "step": 165
+    },
+    {
+      "epoch": 1.4165333333333332,
+      "grad_norm": 0.059243429040783904,
+      "learning_rate": 2e-05,
+      "loss": 0.0838,
+      "step": 166
+    },
+    {
+      "epoch": 1.4250666666666667,
+      "grad_norm": 0.0621476448363388,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 167
+    },
+    {
+      "epoch": 1.4336,
+      "grad_norm": 0.06083867682720169,
+      "learning_rate": 2e-05,
+      "loss": 0.1119,
+      "step": 168
+    },
+    {
+      "epoch": 1.4421333333333333,
+      "grad_norm": 0.09583165335305677,
+      "learning_rate": 2e-05,
+      "loss": 0.1028,
+      "step": 169
+    },
+    {
+      "epoch": 1.4506666666666668,
+      "grad_norm": 0.06411638581314043,
+      "learning_rate": 2e-05,
+      "loss": 0.1181,
+      "step": 170
+    },
+    {
+      "epoch": 1.4592,
+      "grad_norm": 0.05632977234908945,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 171
+    },
+    {
+      "epoch": 1.4677333333333333,
+      "grad_norm": 0.05666068971337413,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 172
+    },
+    {
+      "epoch": 1.4762666666666666,
+      "grad_norm": 0.04502062723807536,
+      "learning_rate": 2e-05,
+      "loss": 0.0588,
+      "step": 173
+    },
+    {
+      "epoch": 1.4848,
+      "grad_norm": 0.05916500176868301,
+      "learning_rate": 2e-05,
+      "loss": 0.0949,
+      "step": 174
+    },
+    {
+      "epoch": 1.4933333333333334,
+      "grad_norm": 0.056484273808864845,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 175
+    },
+    {
+      "epoch": 1.5018666666666667,
+      "grad_norm": 0.06652084448571674,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 176
+    },
+    {
+      "epoch": 1.5104,
+      "grad_norm": 0.05400238078068516,
+      "learning_rate": 2e-05,
+      "loss": 0.0919,
+      "step": 177
+    },
+    {
+      "epoch": 1.5189333333333335,
+      "grad_norm": 0.04979579743346662,
+      "learning_rate": 2e-05,
+      "loss": 0.0879,
+      "step": 178
+    },
+    {
+      "epoch": 1.5274666666666668,
+      "grad_norm": 0.06876105414733971,
+      "learning_rate": 2e-05,
+      "loss": 0.1162,
+      "step": 179
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.05633262015499721,
+      "learning_rate": 2e-05,
+      "loss": 0.1142,
+      "step": 180
+    },
+    {
+      "epoch": 1.5445333333333333,
+      "grad_norm": 0.0599508967519892,
+      "learning_rate": 2e-05,
+      "loss": 0.1073,
+      "step": 181
+    },
+    {
+      "epoch": 1.5530666666666666,
+      "grad_norm": 0.058912170976454126,
+      "learning_rate": 2e-05,
+      "loss": 0.1102,
+      "step": 182
+    },
+    {
+      "epoch": 1.5615999999999999,
+      "grad_norm": 0.05361414988566254,
+      "learning_rate": 2e-05,
+      "loss": 0.0885,
+      "step": 183
+    },
+    {
+      "epoch": 1.5701333333333334,
+      "grad_norm": 0.04517277847384077,
+      "learning_rate": 2e-05,
+      "loss": 0.0763,
+      "step": 184
+    },
+    {
+      "epoch": 1.5786666666666667,
+      "grad_norm": 0.05551553662051771,
+      "learning_rate": 2e-05,
+      "loss": 0.0877,
+      "step": 185
+    },
+    {
+      "epoch": 1.5872000000000002,
+      "grad_norm": 0.05814223969236194,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 186
+    },
+    {
+      "epoch": 1.5957333333333334,
+      "grad_norm": 0.05710054166191597,
+      "learning_rate": 2e-05,
+      "loss": 0.0962,
+      "step": 187
+    },
+    {
+      "epoch": 1.6042666666666667,
+      "grad_norm": 0.054744343932104075,
+      "learning_rate": 2e-05,
+      "loss": 0.0873,
+      "step": 188
+    },
+    {
+      "epoch": 1.6128,
+      "grad_norm": 0.051145521687090995,
+      "learning_rate": 2e-05,
+      "loss": 0.0855,
+      "step": 189
+    },
+    {
+      "epoch": 1.6213333333333333,
+      "grad_norm": 0.05414658860737789,
+      "learning_rate": 2e-05,
+      "loss": 0.0872,
+      "step": 190
+    },
+    {
+      "epoch": 1.6298666666666666,
+      "grad_norm": 0.05084744595533029,
+      "learning_rate": 2e-05,
+      "loss": 0.0891,
+      "step": 191
+    },
+    {
+      "epoch": 1.6383999999999999,
+      "grad_norm": 0.0567070082820597,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 192
+    },
+    {
+      "epoch": 1.6469333333333334,
+      "grad_norm": 0.0494785311411315,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 193
+    },
+    {
+      "epoch": 1.6554666666666666,
+      "grad_norm": 0.062341158530385396,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 194
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.059888336716275685,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 195
+    },
+    {
+      "epoch": 1.6725333333333334,
+      "grad_norm": 0.07346562318829057,
+      "learning_rate": 2e-05,
+      "loss": 0.1329,
+      "step": 196
+    },
+    {
+      "epoch": 1.6810666666666667,
+      "grad_norm": 0.0792360016934733,
+      "learning_rate": 2e-05,
+      "loss": 0.1392,
+      "step": 197
+    },
+    {
+      "epoch": 1.6896,
+      "grad_norm": 0.0800342963229883,
+      "learning_rate": 2e-05,
+      "loss": 0.1199,
+      "step": 198
+    },
+    {
+      "epoch": 1.6981333333333333,
+      "grad_norm": 0.06848045933195548,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 199
+    },
+    {
+      "epoch": 1.7066666666666666,
+      "grad_norm": 0.05743199019316764,
+      "learning_rate": 2e-05,
+      "loss": 0.0811,
+      "step": 200
+    },
+    {
+      "epoch": 1.7151999999999998,
+      "grad_norm": 0.07170531168284446,
+      "learning_rate": 2e-05,
+      "loss": 0.1079,
+      "step": 201
+    },
+    {
+      "epoch": 1.7237333333333333,
+      "grad_norm": 0.05772905481368506,
+      "learning_rate": 2e-05,
+      "loss": 0.0844,
+      "step": 202
+    },
+    {
+      "epoch": 1.7322666666666666,
+      "grad_norm": 0.07504946014098464,
+      "learning_rate": 2e-05,
+      "loss": 0.1257,
+      "step": 203
+    },
+    {
+      "epoch": 1.7408000000000001,
+      "grad_norm": 0.06450179819785838,
+      "learning_rate": 2e-05,
+      "loss": 0.1104,
+      "step": 204
+    },
+    {
+      "epoch": 1.7493333333333334,
+      "grad_norm": 0.06595445965110332,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 205
+    },
+    {
+      "epoch": 1.7578666666666667,
+      "grad_norm": 0.07203558121131749,
+      "learning_rate": 2e-05,
+      "loss": 0.1117,
+      "step": 206
+    },
+    {
+      "epoch": 1.7664,
+      "grad_norm": 0.05954646782409283,
+      "learning_rate": 2e-05,
+      "loss": 0.0729,
+      "step": 207
+    },
+    {
+      "epoch": 1.7749333333333333,
+      "grad_norm": 0.06624894584410884,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 208
+    },
+    {
+      "epoch": 1.7834666666666665,
+      "grad_norm": 0.06888562028256219,
+      "learning_rate": 2e-05,
+      "loss": 0.1398,
+      "step": 209
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.061224088077794406,
+      "learning_rate": 2e-05,
+      "loss": 0.1112,
+      "step": 210
+    },
+    {
+      "epoch": 1.8005333333333333,
+      "grad_norm": 0.06857358883856608,
+      "learning_rate": 2e-05,
+      "loss": 0.1293,
+      "step": 211
+    },
+    {
+      "epoch": 1.8090666666666668,
+      "grad_norm": 0.06177352416779139,
+      "learning_rate": 2e-05,
+      "loss": 0.0884,
+      "step": 212
+    },
+    {
+      "epoch": 1.8176,
+      "grad_norm": 0.08324567429925228,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 213
+    },
+    {
+      "epoch": 1.8261333333333334,
+      "grad_norm": 0.06771677297787752,
+      "learning_rate": 2e-05,
+      "loss": 0.089,
+      "step": 214
+    },
+    {
+      "epoch": 1.8346666666666667,
+      "grad_norm": 0.07055754809472485,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 215
+    },
+    {
+      "epoch": 1.8432,
+      "grad_norm": 0.05856797724392531,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 216
+    },
+    {
+      "epoch": 1.8517333333333332,
+      "grad_norm": 0.07555286129801597,
+      "learning_rate": 2e-05,
+      "loss": 0.0913,
+      "step": 217
+    },
+    {
+      "epoch": 1.8602666666666665,
+      "grad_norm": 0.09242462538643775,
+      "learning_rate": 2e-05,
+      "loss": 0.1241,
+      "step": 218
+    },
+    {
+      "epoch": 1.8688,
+      "grad_norm": 0.06550805537088374,
+      "learning_rate": 2e-05,
+      "loss": 0.0819,
+      "step": 219
+    },
+    {
+      "epoch": 1.8773333333333333,
+      "grad_norm": 0.06016048263236861,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 220
+    },
+    {
+      "epoch": 1.8858666666666668,
+      "grad_norm": 0.06856661106001445,
+      "learning_rate": 2e-05,
+      "loss": 0.1132,
+      "step": 221
+    },
+    {
+      "epoch": 1.8944,
+      "grad_norm": 0.06383306455000529,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 222
+    },
+    {
+      "epoch": 1.9029333333333334,
+      "grad_norm": 0.07240472757239624,
+      "learning_rate": 2e-05,
+      "loss": 0.0863,
+      "step": 223
+    },
+    {
+      "epoch": 1.9114666666666666,
+      "grad_norm": 0.07845654471077741,
+      "learning_rate": 2e-05,
+      "loss": 0.1284,
+      "step": 224
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.07192185833649212,
+      "learning_rate": 2e-05,
+      "loss": 0.1101,
+      "step": 225
+    },
+    {
+      "epoch": 1.9285333333333332,
+      "grad_norm": 0.06390598005596872,
+      "learning_rate": 2e-05,
+      "loss": 0.0917,
+      "step": 226
+    },
+    {
+      "epoch": 1.9370666666666667,
+      "grad_norm": 0.06306138712432224,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 227
+    },
+    {
+      "epoch": 1.9456,
+      "grad_norm": 0.06771381941296478,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 228
+    },
+    {
+      "epoch": 1.9541333333333335,
+      "grad_norm": 0.05899006803461524,
+      "learning_rate": 2e-05,
+      "loss": 0.0782,
+      "step": 229
+    },
+    {
+      "epoch": 1.9626666666666668,
+      "grad_norm": 0.07426956281950735,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 230
+    },
+    {
+      "epoch": 1.9712,
+      "grad_norm": 0.06567534902293475,
+      "learning_rate": 2e-05,
+      "loss": 0.0956,
+      "step": 231
+    },
+    {
+      "epoch": 1.9797333333333333,
+      "grad_norm": 0.07430395142282198,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 232
+    },
+    {
+      "epoch": 1.9882666666666666,
+      "grad_norm": 0.05834447367264806,
+      "learning_rate": 2e-05,
+      "loss": 0.0767,
+      "step": 233
+    },
+    {
+      "epoch": 1.9968,
+      "grad_norm": 0.07137090413877054,
+      "learning_rate": 2e-05,
+      "loss": 0.0821,
+      "step": 234
+    },
+    {
+      "epoch": 2.005333333333333,
+      "grad_norm": 0.07797914240805551,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 235
+    },
+    {
+      "epoch": 2.0138666666666665,
+      "grad_norm": 0.09335648879374187,
+      "learning_rate": 2e-05,
+      "loss": 0.1632,
+      "step": 236
+    },
+    {
+      "epoch": 2.0224,
+      "grad_norm": 0.08280719749100944,
+      "learning_rate": 2e-05,
+      "loss": 0.115,
+      "step": 237
+    },
+    {
+      "epoch": 2.0309333333333335,
+      "grad_norm": 0.08956213539053312,
+      "learning_rate": 2e-05,
+      "loss": 0.109,
+      "step": 238
+    },
+    {
+      "epoch": 2.0394666666666668,
+      "grad_norm": 0.08849786687188893,
+      "learning_rate": 2e-05,
+      "loss": 0.1234,
+      "step": 239
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.16956065138312856,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 240
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 351,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.4262663119891333e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-240/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa4edad1da5557fcd52a6da980443588016cead6f0444a3562cfa68029c66a04
+size 6840

checkpoint-240/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

checkpoint-260/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ../ckpts/Meta-Llama-3-8B-Instruct
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.11.1

checkpoint-260/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "../ckpts/Meta-Llama-3-8B-Instruct",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "gate_proj",
+    "q_proj",
+    "lm_head",
+    "o_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-260/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:eec3bb4f61fc37676975013d46cc5e10f8d9a4b327e7ac49c4d0638301be3de0
+size 1138856856

checkpoint-260/global_step260/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:56b0e33a2082f3838f1e171c77e1243d6f245ca1af4e631b50d6a771d4657e80
+size 528781328

checkpoint-260/global_step260/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2feae52818e8eefd4438e8c94312d2a2181070edb069f9c232d0ad7481fe86c6
+size 199905337

checkpoint-260/latest

@@ -0,0 +1 @@
+global_step260

checkpoint-260/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:df5336a6bdf866da6f2d48b2de7a869897f2fbda5c5e2f70f21996192cb79dbd
+size 14244

checkpoint-260/trainer_state.json

@@ -0,0 +1,1853 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.2186666666666666,
+  "eval_steps": 500,
+  "global_step": 260,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008533333333333334,
+      "grad_norm": 100.21848203113535,
+      "learning_rate": 0.0,
+      "loss": 7.1962,
+      "step": 1
+    },
+    {
+      "epoch": 0.017066666666666667,
+      "grad_norm": 96.43006188910957,
+      "learning_rate": 3.0102999566398115e-06,
+      "loss": 6.9414,
+      "step": 2
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 97.35803466618715,
+      "learning_rate": 4.771212547196624e-06,
+      "loss": 7.0102,
+      "step": 3
+    },
+    {
+      "epoch": 0.034133333333333335,
+      "grad_norm": 95.14837816372646,
+      "learning_rate": 6.020599913279623e-06,
+      "loss": 6.5295,
+      "step": 4
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 91.76544275692784,
+      "learning_rate": 6.989700043360187e-06,
+      "loss": 6.4806,
+      "step": 5
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 84.4494318688335,
+      "learning_rate": 7.781512503836437e-06,
+      "loss": 6.4194,
+      "step": 6
+    },
+    {
+      "epoch": 0.05973333333333333,
+      "grad_norm": 71.37977490595638,
+      "learning_rate": 8.450980400142568e-06,
+      "loss": 5.4953,
+      "step": 7
+    },
+    {
+      "epoch": 0.06826666666666667,
+      "grad_norm": 49.31153456754566,
+      "learning_rate": 9.030899869919434e-06,
+      "loss": 5.4123,
+      "step": 8
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 20.37296364560341,
+      "learning_rate": 9.542425094393249e-06,
+      "loss": 5.2334,
+      "step": 9
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 7.968467079076881,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 5.0282,
+      "step": 10
+    },
+    {
+      "epoch": 0.09386666666666667,
+      "grad_norm": 3.559446532055649,
+      "learning_rate": 1.041392685158225e-05,
+      "loss": 4.612,
+      "step": 11
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 3.5528846947995674,
+      "learning_rate": 1.0791812460476248e-05,
+      "loss": 4.9475,
+      "step": 12
+    },
+    {
+      "epoch": 0.11093333333333333,
+      "grad_norm": 3.541968897471334,
+      "learning_rate": 1.1139433523068365e-05,
+      "loss": 4.2777,
+      "step": 13
+    },
+    {
+      "epoch": 0.11946666666666667,
+      "grad_norm": 3.54718070036198,
+      "learning_rate": 1.1461280356782378e-05,
+      "loss": 4.3507,
+      "step": 14
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 3.8632334830606747,
+      "learning_rate": 1.1760912590556813e-05,
+      "loss": 4.5364,
+      "step": 15
+    },
+    {
+      "epoch": 0.13653333333333334,
+      "grad_norm": 3.6637424744054004,
+      "learning_rate": 1.2041199826559246e-05,
+      "loss": 3.9672,
+      "step": 16
+    },
+    {
+      "epoch": 0.14506666666666668,
+      "grad_norm": 3.919802904818311,
+      "learning_rate": 1.230448921378274e-05,
+      "loss": 4.0618,
+      "step": 17
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 4.71904950738746,
+      "learning_rate": 1.2552725051033058e-05,
+      "loss": 4.6656,
+      "step": 18
+    },
+    {
+      "epoch": 0.16213333333333332,
+      "grad_norm": 4.6656317698690835,
+      "learning_rate": 1.2787536009528288e-05,
+      "loss": 4.1131,
+      "step": 19
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 5.145138692367417,
+      "learning_rate": 1.301029995663981e-05,
+      "loss": 4.0989,
+      "step": 20
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 5.923538014759818,
+      "learning_rate": 1.3222192947339192e-05,
+      "loss": 4.4991,
+      "step": 21
+    },
+    {
+      "epoch": 0.18773333333333334,
+      "grad_norm": 5.941056962941364,
+      "learning_rate": 1.3424226808222062e-05,
+      "loss": 4.0836,
+      "step": 22
+    },
+    {
+      "epoch": 0.19626666666666667,
+      "grad_norm": 6.171026012117947,
+      "learning_rate": 1.3617278360175927e-05,
+      "loss": 3.6861,
+      "step": 23
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 7.130542138930838,
+      "learning_rate": 1.380211241711606e-05,
+      "loss": 4.0958,
+      "step": 24
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 7.328837606110418,
+      "learning_rate": 1.3979400086720374e-05,
+      "loss": 3.9524,
+      "step": 25
+    },
+    {
+      "epoch": 0.22186666666666666,
+      "grad_norm": 6.923489005711429,
+      "learning_rate": 1.4149733479708178e-05,
+      "loss": 3.6062,
+      "step": 26
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 6.409498926059221,
+      "learning_rate": 1.4313637641589872e-05,
+      "loss": 3.2034,
+      "step": 27
+    },
+    {
+      "epoch": 0.23893333333333333,
+      "grad_norm": 5.781628405584682,
+      "learning_rate": 1.4471580313422191e-05,
+      "loss": 2.8158,
+      "step": 28
+    },
+    {
+      "epoch": 0.24746666666666667,
+      "grad_norm": 6.2927590068243315,
+      "learning_rate": 1.4623979978989559e-05,
+      "loss": 2.9803,
+      "step": 29
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 6.103844678630006,
+      "learning_rate": 1.4771212547196623e-05,
+      "loss": 2.847,
+      "step": 30
+    },
+    {
+      "epoch": 0.26453333333333334,
+      "grad_norm": 7.656341925867144,
+      "learning_rate": 1.4913616938342726e-05,
+      "loss": 3.0907,
+      "step": 31
+    },
+    {
+      "epoch": 0.2730666666666667,
+      "grad_norm": 6.324242877501844,
+      "learning_rate": 1.5051499783199059e-05,
+      "loss": 2.3467,
+      "step": 32
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 7.606313101162739,
+      "learning_rate": 1.5185139398778874e-05,
+      "loss": 2.5292,
+      "step": 33
+    },
+    {
+      "epoch": 0.29013333333333335,
+      "grad_norm": 8.553792493849265,
+      "learning_rate": 1.531478917042255e-05,
+      "loss": 2.4547,
+      "step": 34
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 8.483368703272543,
+      "learning_rate": 1.5440680443502753e-05,
+      "loss": 2.1956,
+      "step": 35
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 8.364739352838077,
+      "learning_rate": 1.5563025007672873e-05,
+      "loss": 1.8552,
+      "step": 36
+    },
+    {
+      "epoch": 0.3157333333333333,
+      "grad_norm": 9.37663682000104,
+      "learning_rate": 1.5682017240669948e-05,
+      "loss": 1.9228,
+      "step": 37
+    },
+    {
+      "epoch": 0.32426666666666665,
+      "grad_norm": 9.823047193440066,
+      "learning_rate": 1.57978359661681e-05,
+      "loss": 1.7033,
+      "step": 38
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 9.692618512955894,
+      "learning_rate": 1.591064607026499e-05,
+      "loss": 1.3768,
+      "step": 39
+    },
+    {
+      "epoch": 0.3413333333333333,
+      "grad_norm": 9.08889021911031,
+      "learning_rate": 1.6020599913279622e-05,
+      "loss": 1.3015,
+      "step": 40
+    },
+    {
+      "epoch": 0.34986666666666666,
+      "grad_norm": 8.081534221516058,
+      "learning_rate": 1.6127838567197353e-05,
+      "loss": 0.9228,
+      "step": 41
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 6.238638048950311,
+      "learning_rate": 1.6232492903979005e-05,
+      "loss": 0.7267,
+      "step": 42
+    },
+    {
+      "epoch": 0.36693333333333333,
+      "grad_norm": 3.4058036861773604,
+      "learning_rate": 1.6334684555795865e-05,
+      "loss": 0.5875,
+      "step": 43
+    },
+    {
+      "epoch": 0.37546666666666667,
+      "grad_norm": 2.079163829467713,
+      "learning_rate": 1.6434526764861872e-05,
+      "loss": 0.6355,
+      "step": 44
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 1.5597487833024746,
+      "learning_rate": 1.6532125137753435e-05,
+      "loss": 0.5106,
+      "step": 45
+    },
+    {
+      "epoch": 0.39253333333333335,
+      "grad_norm": 2.491689602375256,
+      "learning_rate": 1.662757831681574e-05,
+      "loss": 0.6454,
+      "step": 46
+    },
+    {
+      "epoch": 0.4010666666666667,
+      "grad_norm": 2.010880438195854,
+      "learning_rate": 1.672097857935717e-05,
+      "loss": 0.4757,
+      "step": 47
+    },
+    {
+      "epoch": 0.4096,
+      "grad_norm": 1.9452805114322096,
+      "learning_rate": 1.681241237375587e-05,
+      "loss": 0.4133,
+      "step": 48
+    },
+    {
+      "epoch": 0.41813333333333336,
+      "grad_norm": 1.7620848552306103,
+      "learning_rate": 1.6901960800285137e-05,
+      "loss": 0.4004,
+      "step": 49
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 1.278224489774809,
+      "learning_rate": 1.6989700043360187e-05,
+      "loss": 0.3523,
+      "step": 50
+    },
+    {
+      "epoch": 0.4352,
+      "grad_norm": 1.6151354758303231,
+      "learning_rate": 1.7075701760979363e-05,
+      "loss": 0.4317,
+      "step": 51
+    },
+    {
+      "epoch": 0.4437333333333333,
+      "grad_norm": 1.3451396055695035,
+      "learning_rate": 1.716003343634799e-05,
+      "loss": 0.3474,
+      "step": 52
+    },
+    {
+      "epoch": 0.45226666666666665,
+      "grad_norm": 1.6814977782362666,
+      "learning_rate": 1.724275869600789e-05,
+      "loss": 0.3706,
+      "step": 53
+    },
+    {
+      "epoch": 0.4608,
+      "grad_norm": 1.1682442432688667,
+      "learning_rate": 1.7323937598229687e-05,
+      "loss": 0.3488,
+      "step": 54
+    },
+    {
+      "epoch": 0.4693333333333333,
+      "grad_norm": 0.8839814540462471,
+      "learning_rate": 1.7403626894942437e-05,
+      "loss": 0.293,
+      "step": 55
+    },
+    {
+      "epoch": 0.47786666666666666,
+      "grad_norm": 0.7974304806295485,
+      "learning_rate": 1.7481880270062003e-05,
+      "loss": 0.2717,
+      "step": 56
+    },
+    {
+      "epoch": 0.4864,
+      "grad_norm": 0.6232203657585239,
+      "learning_rate": 1.7558748556724913e-05,
+      "loss": 0.1741,
+      "step": 57
+    },
+    {
+      "epoch": 0.49493333333333334,
+      "grad_norm": 0.6850688604563008,
+      "learning_rate": 1.763427993562937e-05,
+      "loss": 0.228,
+      "step": 58
+    },
+    {
+      "epoch": 0.5034666666666666,
+      "grad_norm": 0.5923826384300431,
+      "learning_rate": 1.7708520116421443e-05,
+      "loss": 0.2131,
+      "step": 59
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.489754430485032,
+      "learning_rate": 1.7781512503836432e-05,
+      "loss": 0.165,
+      "step": 60
+    },
+    {
+      "epoch": 0.5205333333333333,
+      "grad_norm": 0.5280940052395061,
+      "learning_rate": 1.7853298350107667e-05,
+      "loss": 0.1658,
+      "step": 61
+    },
+    {
+      "epoch": 0.5290666666666667,
+      "grad_norm": 0.4750905992036739,
+      "learning_rate": 1.7923916894982537e-05,
+      "loss": 0.1438,
+      "step": 62
+    },
+    {
+      "epoch": 0.5376,
+      "grad_norm": 0.4162998392722401,
+      "learning_rate": 1.7993405494535815e-05,
+      "loss": 0.1555,
+      "step": 63
+    },
+    {
+      "epoch": 0.5461333333333334,
+      "grad_norm": 0.26863266246370443,
+      "learning_rate": 1.806179973983887e-05,
+      "loss": 0.1323,
+      "step": 64
+    },
+    {
+      "epoch": 0.5546666666666666,
+      "grad_norm": 0.26534221125601215,
+      "learning_rate": 1.8129133566428553e-05,
+      "loss": 0.1671,
+      "step": 65
+    },
+    {
+      "epoch": 0.5632,
+      "grad_norm": 0.2548662962257576,
+      "learning_rate": 1.8195439355418686e-05,
+      "loss": 0.1308,
+      "step": 66
+    },
+    {
+      "epoch": 0.5717333333333333,
+      "grad_norm": 0.18045394638578796,
+      "learning_rate": 1.8260748027008263e-05,
+      "loss": 0.1262,
+      "step": 67
+    },
+    {
+      "epoch": 0.5802666666666667,
+      "grad_norm": 0.17070388073612064,
+      "learning_rate": 1.8325089127062364e-05,
+      "loss": 0.1192,
+      "step": 68
+    },
+    {
+      "epoch": 0.5888,
+      "grad_norm": 0.1531381679244776,
+      "learning_rate": 1.8388490907372553e-05,
+      "loss": 0.1274,
+      "step": 69
+    },
+    {
+      "epoch": 0.5973333333333334,
+      "grad_norm": 0.18196434993687946,
+      "learning_rate": 1.8450980400142568e-05,
+      "loss": 0.1375,
+      "step": 70
+    },
+    {
+      "epoch": 0.6058666666666667,
+      "grad_norm": 0.15324416972951205,
+      "learning_rate": 1.8512583487190752e-05,
+      "loss": 0.1599,
+      "step": 71
+    },
+    {
+      "epoch": 0.6144,
+      "grad_norm": 0.10884462064503801,
+      "learning_rate": 1.857332496431268e-05,
+      "loss": 0.1041,
+      "step": 72
+    },
+    {
+      "epoch": 0.6229333333333333,
+      "grad_norm": 0.12915133528192668,
+      "learning_rate": 1.8633228601204555e-05,
+      "loss": 0.1406,
+      "step": 73
+    },
+    {
+      "epoch": 0.6314666666666666,
+      "grad_norm": 0.12553425699952878,
+      "learning_rate": 1.8692317197309763e-05,
+      "loss": 0.1256,
+      "step": 74
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11976960918968543,
+      "learning_rate": 1.8750612633916997e-05,
+      "loss": 0.1144,
+      "step": 75
+    },
+    {
+      "epoch": 0.6485333333333333,
+      "grad_norm": 0.115805998298789,
+      "learning_rate": 1.8808135922807914e-05,
+      "loss": 0.1528,
+      "step": 76
+    },
+    {
+      "epoch": 0.6570666666666667,
+      "grad_norm": 0.10325948496697443,
+      "learning_rate": 1.8864907251724818e-05,
+      "loss": 0.1044,
+      "step": 77
+    },
+    {
+      "epoch": 0.6656,
+      "grad_norm": 0.09595064346541006,
+      "learning_rate": 1.8920946026904802e-05,
+      "loss": 0.1534,
+      "step": 78
+    },
+    {
+      "epoch": 0.6741333333333334,
+      "grad_norm": 0.08796742845240496,
+      "learning_rate": 1.8976270912904414e-05,
+      "loss": 0.1155,
+      "step": 79
+    },
+    {
+      "epoch": 0.6826666666666666,
+      "grad_norm": 0.08218991738379527,
+      "learning_rate": 1.9030899869919434e-05,
+      "loss": 0.1311,
+      "step": 80
+    },
+    {
+      "epoch": 0.6912,
+      "grad_norm": 0.08290815261109215,
+      "learning_rate": 1.9084850188786497e-05,
+      "loss": 0.11,
+      "step": 81
+    },
+    {
+      "epoch": 0.6997333333333333,
+      "grad_norm": 0.0794132180176064,
+      "learning_rate": 1.9138138523837165e-05,
+      "loss": 0.1135,
+      "step": 82
+    },
+    {
+      "epoch": 0.7082666666666667,
+      "grad_norm": 0.06934410705255296,
+      "learning_rate": 1.919078092376074e-05,
+      "loss": 0.109,
+      "step": 83
+    },
+    {
+      "epoch": 0.7168,
+      "grad_norm": 0.09000563031870593,
+      "learning_rate": 1.9242792860618813e-05,
+      "loss": 0.12,
+      "step": 84
+    },
+    {
+      "epoch": 0.7253333333333334,
+      "grad_norm": 0.1134042277879818,
+      "learning_rate": 1.929418925714293e-05,
+      "loss": 0.1223,
+      "step": 85
+    },
+    {
+      "epoch": 0.7338666666666667,
+      "grad_norm": 0.09118764690233076,
+      "learning_rate": 1.9344984512435673e-05,
+      "loss": 0.1459,
+      "step": 86
+    },
+    {
+      "epoch": 0.7424,
+      "grad_norm": 0.07873016754353963,
+      "learning_rate": 1.9395192526186183e-05,
+      "loss": 0.1422,
+      "step": 87
+    },
+    {
+      "epoch": 0.7509333333333333,
+      "grad_norm": 0.1796495874463076,
+      "learning_rate": 1.9444826721501687e-05,
+      "loss": 0.1291,
+      "step": 88
+    },
+    {
+      "epoch": 0.7594666666666666,
+      "grad_norm": 0.0679589944174269,
+      "learning_rate": 1.9493900066449125e-05,
+      "loss": 0.108,
+      "step": 89
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.08174688574235538,
+      "learning_rate": 1.9542425094393246e-05,
+      "loss": 0.1081,
+      "step": 90
+    },
+    {
+      "epoch": 0.7765333333333333,
+      "grad_norm": 0.057137370501406756,
+      "learning_rate": 1.9590413923210934e-05,
+      "loss": 0.0934,
+      "step": 91
+    },
+    {
+      "epoch": 0.7850666666666667,
+      "grad_norm": 0.06578111924908255,
+      "learning_rate": 1.9637878273455555e-05,
+      "loss": 0.1085,
+      "step": 92
+    },
+    {
+      "epoch": 0.7936,
+      "grad_norm": 0.08945990540906254,
+      "learning_rate": 1.968482948553935e-05,
+      "loss": 0.1747,
+      "step": 93
+    },
+    {
+      "epoch": 0.8021333333333334,
+      "grad_norm": 0.06183863311044229,
+      "learning_rate": 1.9731278535996986e-05,
+      "loss": 0.1136,
+      "step": 94
+    },
+    {
+      "epoch": 0.8106666666666666,
+      "grad_norm": 0.05777899602544702,
+      "learning_rate": 1.9777236052888476e-05,
+      "loss": 0.0984,
+      "step": 95
+    },
+    {
+      "epoch": 0.8192,
+      "grad_norm": 0.08130851607693534,
+      "learning_rate": 1.9822712330395683e-05,
+      "loss": 0.187,
+      "step": 96
+    },
+    {
+      "epoch": 0.8277333333333333,
+      "grad_norm": 0.06426546202002927,
+      "learning_rate": 1.986771734266245e-05,
+      "loss": 0.1296,
+      "step": 97
+    },
+    {
+      "epoch": 0.8362666666666667,
+      "grad_norm": 0.069692313707994,
+      "learning_rate": 1.991226075692495e-05,
+      "loss": 0.1404,
+      "step": 98
+    },
+    {
+      "epoch": 0.8448,
+      "grad_norm": 0.05494542266886729,
+      "learning_rate": 1.9956351945975496e-05,
+      "loss": 0.116,
+      "step": 99
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.07571686966840627,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 0.1539,
+      "step": 100
+    },
+    {
+      "epoch": 0.8618666666666667,
+      "grad_norm": 0.054351059117603705,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 101
+    },
+    {
+      "epoch": 0.8704,
+      "grad_norm": 0.06531899611551092,
+      "learning_rate": 2e-05,
+      "loss": 0.0827,
+      "step": 102
+    },
+    {
+      "epoch": 0.8789333333333333,
+      "grad_norm": 0.06131678646504652,
+      "learning_rate": 2e-05,
+      "loss": 0.1266,
+      "step": 103
+    },
+    {
+      "epoch": 0.8874666666666666,
+      "grad_norm": 0.06850220540661824,
+      "learning_rate": 2e-05,
+      "loss": 0.1456,
+      "step": 104
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.05806908951252483,
+      "learning_rate": 2e-05,
+      "loss": 0.0954,
+      "step": 105
+    },
+    {
+      "epoch": 0.9045333333333333,
+      "grad_norm": 0.06503642452033717,
+      "learning_rate": 2e-05,
+      "loss": 0.1417,
+      "step": 106
+    },
+    {
+      "epoch": 0.9130666666666667,
+      "grad_norm": 0.050486271853277066,
+      "learning_rate": 2e-05,
+      "loss": 0.0959,
+      "step": 107
+    },
+    {
+      "epoch": 0.9216,
+      "grad_norm": 0.07746063813802379,
+      "learning_rate": 2e-05,
+      "loss": 0.1256,
+      "step": 108
+    },
+    {
+      "epoch": 0.9301333333333334,
+      "grad_norm": 0.051231172380840004,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 109
+    },
+    {
+      "epoch": 0.9386666666666666,
+      "grad_norm": 0.056296443557859455,
+      "learning_rate": 2e-05,
+      "loss": 0.1056,
+      "step": 110
+    },
+    {
+      "epoch": 0.9472,
+      "grad_norm": 0.05058663240713958,
+      "learning_rate": 2e-05,
+      "loss": 0.0971,
+      "step": 111
+    },
+    {
+      "epoch": 0.9557333333333333,
+      "grad_norm": 0.05532886570130611,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 112
+    },
+    {
+      "epoch": 0.9642666666666667,
+      "grad_norm": 0.05327811326654907,
+      "learning_rate": 2e-05,
+      "loss": 0.0989,
+      "step": 113
+    },
+    {
+      "epoch": 0.9728,
+      "grad_norm": 0.05663279364147864,
+      "learning_rate": 2e-05,
+      "loss": 0.0958,
+      "step": 114
+    },
+    {
+      "epoch": 0.9813333333333333,
+      "grad_norm": 0.04930904541225805,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 115
+    },
+    {
+      "epoch": 0.9898666666666667,
+      "grad_norm": 0.06096947951115022,
+      "learning_rate": 2e-05,
+      "loss": 0.106,
+      "step": 116
+    },
+    {
+      "epoch": 0.9984,
+      "grad_norm": 0.050092322361182495,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 117
+    },
+    {
+      "epoch": 1.0069333333333332,
+      "grad_norm": 0.04980408443758999,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 118
+    },
+    {
+      "epoch": 1.0154666666666667,
+      "grad_norm": 0.051183082721834305,
+      "learning_rate": 2e-05,
+      "loss": 0.1049,
+      "step": 119
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.04332220265802814,
+      "learning_rate": 2e-05,
+      "loss": 0.0983,
+      "step": 120
+    },
+    {
+      "epoch": 1.0325333333333333,
+      "grad_norm": 0.07211731499677299,
+      "learning_rate": 2e-05,
+      "loss": 0.1386,
+      "step": 121
+    },
+    {
+      "epoch": 1.0410666666666666,
+      "grad_norm": 0.06550870223740553,
+      "learning_rate": 2e-05,
+      "loss": 0.1334,
+      "step": 122
+    },
+    {
+      "epoch": 1.0496,
+      "grad_norm": 0.05331839690767287,
+      "learning_rate": 2e-05,
+      "loss": 0.1014,
+      "step": 123
+    },
+    {
+      "epoch": 1.0581333333333334,
+      "grad_norm": 0.05227685628767905,
+      "learning_rate": 2e-05,
+      "loss": 0.1098,
+      "step": 124
+    },
+    {
+      "epoch": 1.0666666666666667,
+      "grad_norm": 0.07641693882491171,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 125
+    },
+    {
+      "epoch": 1.0752,
+      "grad_norm": 0.052835367770791786,
+      "learning_rate": 2e-05,
+      "loss": 0.1231,
+      "step": 126
+    },
+    {
+      "epoch": 1.0837333333333334,
+      "grad_norm": 0.07520329755025788,
+      "learning_rate": 2e-05,
+      "loss": 0.085,
+      "step": 127
+    },
+    {
+      "epoch": 1.0922666666666667,
+      "grad_norm": 0.07670066152157425,
+      "learning_rate": 2e-05,
+      "loss": 0.1071,
+      "step": 128
+    },
+    {
+      "epoch": 1.1008,
+      "grad_norm": 0.052832906560645154,
+      "learning_rate": 2e-05,
+      "loss": 0.1093,
+      "step": 129
+    },
+    {
+      "epoch": 1.1093333333333333,
+      "grad_norm": 0.06573889037311398,
+      "learning_rate": 2e-05,
+      "loss": 0.1193,
+      "step": 130
+    },
+    {
+      "epoch": 1.1178666666666666,
+      "grad_norm": 0.05175471296566334,
+      "learning_rate": 2e-05,
+      "loss": 0.1184,
+      "step": 131
+    },
+    {
+      "epoch": 1.1264,
+      "grad_norm": 0.05912231419793496,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 132
+    },
+    {
+      "epoch": 1.1349333333333333,
+      "grad_norm": 0.04899140475981105,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 133
+    },
+    {
+      "epoch": 1.1434666666666666,
+      "grad_norm": 0.05939916939142137,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 134
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.0516819217599706,
+      "learning_rate": 2e-05,
+      "loss": 0.0834,
+      "step": 135
+    },
+    {
+      "epoch": 1.1605333333333334,
+      "grad_norm": 0.05456440346454737,
+      "learning_rate": 2e-05,
+      "loss": 0.1183,
+      "step": 136
+    },
+    {
+      "epoch": 1.1690666666666667,
+      "grad_norm": 0.059906464476343235,
+      "learning_rate": 2e-05,
+      "loss": 0.1048,
+      "step": 137
+    },
+    {
+      "epoch": 1.1776,
+      "grad_norm": 0.0720112680204319,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 138
+    },
+    {
+      "epoch": 1.1861333333333333,
+      "grad_norm": 0.04940202805828527,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 139
+    },
+    {
+      "epoch": 1.1946666666666665,
+      "grad_norm": 0.060088609545130046,
+      "learning_rate": 2e-05,
+      "loss": 0.0952,
+      "step": 140
+    },
+    {
+      "epoch": 1.2032,
+      "grad_norm": 0.04694761423612446,
+      "learning_rate": 2e-05,
+      "loss": 0.0717,
+      "step": 141
+    },
+    {
+      "epoch": 1.2117333333333333,
+      "grad_norm": 0.05628581562512457,
+      "learning_rate": 2e-05,
+      "loss": 0.1062,
+      "step": 142
+    },
+    {
+      "epoch": 1.2202666666666666,
+      "grad_norm": 0.06876420990437652,
+      "learning_rate": 2e-05,
+      "loss": 0.1218,
+      "step": 143
+    },
+    {
+      "epoch": 1.2288000000000001,
+      "grad_norm": 0.058774700501610655,
+      "learning_rate": 2e-05,
+      "loss": 0.1125,
+      "step": 144
+    },
+    {
+      "epoch": 1.2373333333333334,
+      "grad_norm": 0.061855922064341186,
+      "learning_rate": 2e-05,
+      "loss": 0.1295,
+      "step": 145
+    },
+    {
+      "epoch": 1.2458666666666667,
+      "grad_norm": 0.0813047704730138,
+      "learning_rate": 2e-05,
+      "loss": 0.1165,
+      "step": 146
+    },
+    {
+      "epoch": 1.2544,
+      "grad_norm": 0.061374000305305446,
+      "learning_rate": 2e-05,
+      "loss": 0.1094,
+      "step": 147
+    },
+    {
+      "epoch": 1.2629333333333332,
+      "grad_norm": 0.055537169110833,
+      "learning_rate": 2e-05,
+      "loss": 0.1054,
+      "step": 148
+    },
+    {
+      "epoch": 1.2714666666666667,
+      "grad_norm": 0.04423248714119304,
+      "learning_rate": 2e-05,
+      "loss": 0.0841,
+      "step": 149
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.049931966607835034,
+      "learning_rate": 2e-05,
+      "loss": 0.0961,
+      "step": 150
+    },
+    {
+      "epoch": 1.2885333333333333,
+      "grad_norm": 0.06178656953298769,
+      "learning_rate": 2e-05,
+      "loss": 0.0854,
+      "step": 151
+    },
+    {
+      "epoch": 1.2970666666666666,
+      "grad_norm": 0.05783812343287897,
+      "learning_rate": 2e-05,
+      "loss": 0.1141,
+      "step": 152
+    },
+    {
+      "epoch": 1.3056,
+      "grad_norm": 0.048955120400167584,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 153
+    },
+    {
+      "epoch": 1.3141333333333334,
+      "grad_norm": 0.12270174746806978,
+      "learning_rate": 2e-05,
+      "loss": 0.1553,
+      "step": 154
+    },
+    {
+      "epoch": 1.3226666666666667,
+      "grad_norm": 0.06928026959973474,
+      "learning_rate": 2e-05,
+      "loss": 0.1274,
+      "step": 155
+    },
+    {
+      "epoch": 1.3312,
+      "grad_norm": 0.04756100666105405,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 156
+    },
+    {
+      "epoch": 1.3397333333333332,
+      "grad_norm": 0.056054951338196934,
+      "learning_rate": 2e-05,
+      "loss": 0.0831,
+      "step": 157
+    },
+    {
+      "epoch": 1.3482666666666667,
+      "grad_norm": 0.0516990471964239,
+      "learning_rate": 2e-05,
+      "loss": 0.0883,
+      "step": 158
+    },
+    {
+      "epoch": 1.3568,
+      "grad_norm": 0.06011650542069954,
+      "learning_rate": 2e-05,
+      "loss": 0.0938,
+      "step": 159
+    },
+    {
+      "epoch": 1.3653333333333333,
+      "grad_norm": 0.051831307951873976,
+      "learning_rate": 2e-05,
+      "loss": 0.1019,
+      "step": 160
+    },
+    {
+      "epoch": 1.3738666666666668,
+      "grad_norm": 0.0544902702048936,
+      "learning_rate": 2e-05,
+      "loss": 0.0906,
+      "step": 161
+    },
+    {
+      "epoch": 1.3824,
+      "grad_norm": 0.06057617032526,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 162
+    },
+    {
+      "epoch": 1.3909333333333334,
+      "grad_norm": 0.07288058025189605,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 163
+    },
+    {
+      "epoch": 1.3994666666666666,
+      "grad_norm": 0.05639043792084219,
+      "learning_rate": 2e-05,
+      "loss": 0.1031,
+      "step": 164
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.0586469408837505,
+      "learning_rate": 2e-05,
+      "loss": 0.1143,
+      "step": 165
+    },
+    {
+      "epoch": 1.4165333333333332,
+      "grad_norm": 0.059243429040783904,
+      "learning_rate": 2e-05,
+      "loss": 0.0838,
+      "step": 166
+    },
+    {
+      "epoch": 1.4250666666666667,
+      "grad_norm": 0.0621476448363388,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 167
+    },
+    {
+      "epoch": 1.4336,
+      "grad_norm": 0.06083867682720169,
+      "learning_rate": 2e-05,
+      "loss": 0.1119,
+      "step": 168
+    },
+    {
+      "epoch": 1.4421333333333333,
+      "grad_norm": 0.09583165335305677,
+      "learning_rate": 2e-05,
+      "loss": 0.1028,
+      "step": 169
+    },
+    {
+      "epoch": 1.4506666666666668,
+      "grad_norm": 0.06411638581314043,
+      "learning_rate": 2e-05,
+      "loss": 0.1181,
+      "step": 170
+    },
+    {
+      "epoch": 1.4592,
+      "grad_norm": 0.05632977234908945,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 171
+    },
+    {
+      "epoch": 1.4677333333333333,
+      "grad_norm": 0.05666068971337413,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 172
+    },
+    {
+      "epoch": 1.4762666666666666,
+      "grad_norm": 0.04502062723807536,
+      "learning_rate": 2e-05,
+      "loss": 0.0588,
+      "step": 173
+    },
+    {
+      "epoch": 1.4848,
+      "grad_norm": 0.05916500176868301,
+      "learning_rate": 2e-05,
+      "loss": 0.0949,
+      "step": 174
+    },
+    {
+      "epoch": 1.4933333333333334,
+      "grad_norm": 0.056484273808864845,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 175
+    },
+    {
+      "epoch": 1.5018666666666667,
+      "grad_norm": 0.06652084448571674,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 176
+    },
+    {
+      "epoch": 1.5104,
+      "grad_norm": 0.05400238078068516,
+      "learning_rate": 2e-05,
+      "loss": 0.0919,
+      "step": 177
+    },
+    {
+      "epoch": 1.5189333333333335,
+      "grad_norm": 0.04979579743346662,
+      "learning_rate": 2e-05,
+      "loss": 0.0879,
+      "step": 178
+    },
+    {
+      "epoch": 1.5274666666666668,
+      "grad_norm": 0.06876105414733971,
+      "learning_rate": 2e-05,
+      "loss": 0.1162,
+      "step": 179
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.05633262015499721,
+      "learning_rate": 2e-05,
+      "loss": 0.1142,
+      "step": 180
+    },
+    {
+      "epoch": 1.5445333333333333,
+      "grad_norm": 0.0599508967519892,
+      "learning_rate": 2e-05,
+      "loss": 0.1073,
+      "step": 181
+    },
+    {
+      "epoch": 1.5530666666666666,
+      "grad_norm": 0.058912170976454126,
+      "learning_rate": 2e-05,
+      "loss": 0.1102,
+      "step": 182
+    },
+    {
+      "epoch": 1.5615999999999999,
+      "grad_norm": 0.05361414988566254,
+      "learning_rate": 2e-05,
+      "loss": 0.0885,
+      "step": 183
+    },
+    {
+      "epoch": 1.5701333333333334,
+      "grad_norm": 0.04517277847384077,
+      "learning_rate": 2e-05,
+      "loss": 0.0763,
+      "step": 184
+    },
+    {
+      "epoch": 1.5786666666666667,
+      "grad_norm": 0.05551553662051771,
+      "learning_rate": 2e-05,
+      "loss": 0.0877,
+      "step": 185
+    },
+    {
+      "epoch": 1.5872000000000002,
+      "grad_norm": 0.05814223969236194,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 186
+    },
+    {
+      "epoch": 1.5957333333333334,
+      "grad_norm": 0.05710054166191597,
+      "learning_rate": 2e-05,
+      "loss": 0.0962,
+      "step": 187
+    },
+    {
+      "epoch": 1.6042666666666667,
+      "grad_norm": 0.054744343932104075,
+      "learning_rate": 2e-05,
+      "loss": 0.0873,
+      "step": 188
+    },
+    {
+      "epoch": 1.6128,
+      "grad_norm": 0.051145521687090995,
+      "learning_rate": 2e-05,
+      "loss": 0.0855,
+      "step": 189
+    },
+    {
+      "epoch": 1.6213333333333333,
+      "grad_norm": 0.05414658860737789,
+      "learning_rate": 2e-05,
+      "loss": 0.0872,
+      "step": 190
+    },
+    {
+      "epoch": 1.6298666666666666,
+      "grad_norm": 0.05084744595533029,
+      "learning_rate": 2e-05,
+      "loss": 0.0891,
+      "step": 191
+    },
+    {
+      "epoch": 1.6383999999999999,
+      "grad_norm": 0.0567070082820597,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 192
+    },
+    {
+      "epoch": 1.6469333333333334,
+      "grad_norm": 0.0494785311411315,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 193
+    },
+    {
+      "epoch": 1.6554666666666666,
+      "grad_norm": 0.062341158530385396,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 194
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.059888336716275685,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 195
+    },
+    {
+      "epoch": 1.6725333333333334,
+      "grad_norm": 0.07346562318829057,
+      "learning_rate": 2e-05,
+      "loss": 0.1329,
+      "step": 196
+    },
+    {
+      "epoch": 1.6810666666666667,
+      "grad_norm": 0.0792360016934733,
+      "learning_rate": 2e-05,
+      "loss": 0.1392,
+      "step": 197
+    },
+    {
+      "epoch": 1.6896,
+      "grad_norm": 0.0800342963229883,
+      "learning_rate": 2e-05,
+      "loss": 0.1199,
+      "step": 198
+    },
+    {
+      "epoch": 1.6981333333333333,
+      "grad_norm": 0.06848045933195548,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 199
+    },
+    {
+      "epoch": 1.7066666666666666,
+      "grad_norm": 0.05743199019316764,
+      "learning_rate": 2e-05,
+      "loss": 0.0811,
+      "step": 200
+    },
+    {
+      "epoch": 1.7151999999999998,
+      "grad_norm": 0.07170531168284446,
+      "learning_rate": 2e-05,
+      "loss": 0.1079,
+      "step": 201
+    },
+    {
+      "epoch": 1.7237333333333333,
+      "grad_norm": 0.05772905481368506,
+      "learning_rate": 2e-05,
+      "loss": 0.0844,
+      "step": 202
+    },
+    {
+      "epoch": 1.7322666666666666,
+      "grad_norm": 0.07504946014098464,
+      "learning_rate": 2e-05,
+      "loss": 0.1257,
+      "step": 203
+    },
+    {
+      "epoch": 1.7408000000000001,
+      "grad_norm": 0.06450179819785838,
+      "learning_rate": 2e-05,
+      "loss": 0.1104,
+      "step": 204
+    },
+    {
+      "epoch": 1.7493333333333334,
+      "grad_norm": 0.06595445965110332,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 205
+    },
+    {
+      "epoch": 1.7578666666666667,
+      "grad_norm": 0.07203558121131749,
+      "learning_rate": 2e-05,
+      "loss": 0.1117,
+      "step": 206
+    },
+    {
+      "epoch": 1.7664,
+      "grad_norm": 0.05954646782409283,
+      "learning_rate": 2e-05,
+      "loss": 0.0729,
+      "step": 207
+    },
+    {
+      "epoch": 1.7749333333333333,
+      "grad_norm": 0.06624894584410884,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 208
+    },
+    {
+      "epoch": 1.7834666666666665,
+      "grad_norm": 0.06888562028256219,
+      "learning_rate": 2e-05,
+      "loss": 0.1398,
+      "step": 209
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.061224088077794406,
+      "learning_rate": 2e-05,
+      "loss": 0.1112,
+      "step": 210
+    },
+    {
+      "epoch": 1.8005333333333333,
+      "grad_norm": 0.06857358883856608,
+      "learning_rate": 2e-05,
+      "loss": 0.1293,
+      "step": 211
+    },
+    {
+      "epoch": 1.8090666666666668,
+      "grad_norm": 0.06177352416779139,
+      "learning_rate": 2e-05,
+      "loss": 0.0884,
+      "step": 212
+    },
+    {
+      "epoch": 1.8176,
+      "grad_norm": 0.08324567429925228,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 213
+    },
+    {
+      "epoch": 1.8261333333333334,
+      "grad_norm": 0.06771677297787752,
+      "learning_rate": 2e-05,
+      "loss": 0.089,
+      "step": 214
+    },
+    {
+      "epoch": 1.8346666666666667,
+      "grad_norm": 0.07055754809472485,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 215
+    },
+    {
+      "epoch": 1.8432,
+      "grad_norm": 0.05856797724392531,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 216
+    },
+    {
+      "epoch": 1.8517333333333332,
+      "grad_norm": 0.07555286129801597,
+      "learning_rate": 2e-05,
+      "loss": 0.0913,
+      "step": 217
+    },
+    {
+      "epoch": 1.8602666666666665,
+      "grad_norm": 0.09242462538643775,
+      "learning_rate": 2e-05,
+      "loss": 0.1241,
+      "step": 218
+    },
+    {
+      "epoch": 1.8688,
+      "grad_norm": 0.06550805537088374,
+      "learning_rate": 2e-05,
+      "loss": 0.0819,
+      "step": 219
+    },
+    {
+      "epoch": 1.8773333333333333,
+      "grad_norm": 0.06016048263236861,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 220
+    },
+    {
+      "epoch": 1.8858666666666668,
+      "grad_norm": 0.06856661106001445,
+      "learning_rate": 2e-05,
+      "loss": 0.1132,
+      "step": 221
+    },
+    {
+      "epoch": 1.8944,
+      "grad_norm": 0.06383306455000529,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 222
+    },
+    {
+      "epoch": 1.9029333333333334,
+      "grad_norm": 0.07240472757239624,
+      "learning_rate": 2e-05,
+      "loss": 0.0863,
+      "step": 223
+    },
+    {
+      "epoch": 1.9114666666666666,
+      "grad_norm": 0.07845654471077741,
+      "learning_rate": 2e-05,
+      "loss": 0.1284,
+      "step": 224
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.07192185833649212,
+      "learning_rate": 2e-05,
+      "loss": 0.1101,
+      "step": 225
+    },
+    {
+      "epoch": 1.9285333333333332,
+      "grad_norm": 0.06390598005596872,
+      "learning_rate": 2e-05,
+      "loss": 0.0917,
+      "step": 226
+    },
+    {
+      "epoch": 1.9370666666666667,
+      "grad_norm": 0.06306138712432224,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 227
+    },
+    {
+      "epoch": 1.9456,
+      "grad_norm": 0.06771381941296478,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 228
+    },
+    {
+      "epoch": 1.9541333333333335,
+      "grad_norm": 0.05899006803461524,
+      "learning_rate": 2e-05,
+      "loss": 0.0782,
+      "step": 229
+    },
+    {
+      "epoch": 1.9626666666666668,
+      "grad_norm": 0.07426956281950735,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 230
+    },
+    {
+      "epoch": 1.9712,
+      "grad_norm": 0.06567534902293475,
+      "learning_rate": 2e-05,
+      "loss": 0.0956,
+      "step": 231
+    },
+    {
+      "epoch": 1.9797333333333333,
+      "grad_norm": 0.07430395142282198,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 232
+    },
+    {
+      "epoch": 1.9882666666666666,
+      "grad_norm": 0.05834447367264806,
+      "learning_rate": 2e-05,
+      "loss": 0.0767,
+      "step": 233
+    },
+    {
+      "epoch": 1.9968,
+      "grad_norm": 0.07137090413877054,
+      "learning_rate": 2e-05,
+      "loss": 0.0821,
+      "step": 234
+    },
+    {
+      "epoch": 2.005333333333333,
+      "grad_norm": 0.07797914240805551,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 235
+    },
+    {
+      "epoch": 2.0138666666666665,
+      "grad_norm": 0.09335648879374187,
+      "learning_rate": 2e-05,
+      "loss": 0.1632,
+      "step": 236
+    },
+    {
+      "epoch": 2.0224,
+      "grad_norm": 0.08280719749100944,
+      "learning_rate": 2e-05,
+      "loss": 0.115,
+      "step": 237
+    },
+    {
+      "epoch": 2.0309333333333335,
+      "grad_norm": 0.08956213539053312,
+      "learning_rate": 2e-05,
+      "loss": 0.109,
+      "step": 238
+    },
+    {
+      "epoch": 2.0394666666666668,
+      "grad_norm": 0.08849786687188893,
+      "learning_rate": 2e-05,
+      "loss": 0.1234,
+      "step": 239
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.16956065138312856,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 240
+    },
+    {
+      "epoch": 2.0565333333333333,
+      "grad_norm": 0.07377723133015779,
+      "learning_rate": 2e-05,
+      "loss": 0.1068,
+      "step": 241
+    },
+    {
+      "epoch": 2.0650666666666666,
+      "grad_norm": 0.08839651322303756,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 242
+    },
+    {
+      "epoch": 2.0736,
+      "grad_norm": 0.10641135540308218,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 243
+    },
+    {
+      "epoch": 2.082133333333333,
+      "grad_norm": 0.09370442613891326,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 244
+    },
+    {
+      "epoch": 2.0906666666666665,
+      "grad_norm": 0.08988093042206098,
+      "learning_rate": 2e-05,
+      "loss": 0.1201,
+      "step": 245
+    },
+    {
+      "epoch": 2.0992,
+      "grad_norm": 0.07167309696967938,
+      "learning_rate": 2e-05,
+      "loss": 0.0747,
+      "step": 246
+    },
+    {
+      "epoch": 2.1077333333333335,
+      "grad_norm": 0.08822161286881192,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 247
+    },
+    {
+      "epoch": 2.1162666666666667,
+      "grad_norm": 0.08439713613775747,
+      "learning_rate": 2e-05,
+      "loss": 0.1063,
+      "step": 248
+    },
+    {
+      "epoch": 2.1248,
+      "grad_norm": 0.0942232139678121,
+      "learning_rate": 2e-05,
+      "loss": 0.0865,
+      "step": 249
+    },
+    {
+      "epoch": 2.1333333333333333,
+      "grad_norm": 0.1066306844906037,
+      "learning_rate": 2e-05,
+      "loss": 0.1198,
+      "step": 250
+    },
+    {
+      "epoch": 2.1418666666666666,
+      "grad_norm": 0.10435492421714423,
+      "learning_rate": 2e-05,
+      "loss": 0.0968,
+      "step": 251
+    },
+    {
+      "epoch": 2.1504,
+      "grad_norm": 0.10513719348492025,
+      "learning_rate": 2e-05,
+      "loss": 0.1045,
+      "step": 252
+    },
+    {
+      "epoch": 2.158933333333333,
+      "grad_norm": 0.12580299248756027,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 253
+    },
+    {
+      "epoch": 2.167466666666667,
+      "grad_norm": 0.10265943183351584,
+      "learning_rate": 2e-05,
+      "loss": 0.0892,
+      "step": 254
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.09473119803029131,
+      "learning_rate": 2e-05,
+      "loss": 0.0721,
+      "step": 255
+    },
+    {
+      "epoch": 2.1845333333333334,
+      "grad_norm": 0.13420988295622846,
+      "learning_rate": 2e-05,
+      "loss": 0.1034,
+      "step": 256
+    },
+    {
+      "epoch": 2.1930666666666667,
+      "grad_norm": 0.11474592330595303,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 257
+    },
+    {
+      "epoch": 2.2016,
+      "grad_norm": 0.12382914500613644,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 258
+    },
+    {
+      "epoch": 2.2101333333333333,
+      "grad_norm": 0.12984065755657193,
+      "learning_rate": 2e-05,
+      "loss": 0.0888,
+      "step": 259
+    },
+    {
+      "epoch": 2.2186666666666666,
+      "grad_norm": 0.14210964340218168,
+      "learning_rate": 2e-05,
+      "loss": 0.105,
+      "step": 260
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 351,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 3.714350084800905e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-260/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa4edad1da5557fcd52a6da980443588016cead6f0444a3562cfa68029c66a04
+size 6840

checkpoint-260/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

checkpoint-280/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ../ckpts/Meta-Llama-3-8B-Instruct
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.11.1

checkpoint-280/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "../ckpts/Meta-Llama-3-8B-Instruct",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "gate_proj",
+    "q_proj",
+    "lm_head",
+    "o_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-280/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fe8296cd05ef57432e2f5d4dc15e109f03b45f45fe3dd67cd03570bae08eef72
+size 1138856856

checkpoint-280/global_step280/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c03dca38a1e33b28a991ecf7289c7ae7d13049de38fe7cf6054a3b590895c68b
+size 528781328

checkpoint-280/global_step280/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc1abf0ef27009f0ab6ba0a4bad28969109ddce84672932573daa1fa147a5a1d
+size 199905337

checkpoint-280/latest

@@ -0,0 +1 @@
+global_step280

checkpoint-280/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8ff2e7999b56e8e1e58923b2da20098ae572204c232fe0dde3aa400cf65d3cab
+size 14244

checkpoint-280/trainer_state.json

@@ -0,0 +1,1993 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.389333333333333,
+  "eval_steps": 500,
+  "global_step": 280,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008533333333333334,
+      "grad_norm": 100.21848203113535,
+      "learning_rate": 0.0,
+      "loss": 7.1962,
+      "step": 1
+    },
+    {
+      "epoch": 0.017066666666666667,
+      "grad_norm": 96.43006188910957,
+      "learning_rate": 3.0102999566398115e-06,
+      "loss": 6.9414,
+      "step": 2
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 97.35803466618715,
+      "learning_rate": 4.771212547196624e-06,
+      "loss": 7.0102,
+      "step": 3
+    },
+    {
+      "epoch": 0.034133333333333335,
+      "grad_norm": 95.14837816372646,
+      "learning_rate": 6.020599913279623e-06,
+      "loss": 6.5295,
+      "step": 4
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 91.76544275692784,
+      "learning_rate": 6.989700043360187e-06,
+      "loss": 6.4806,
+      "step": 5
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 84.4494318688335,
+      "learning_rate": 7.781512503836437e-06,
+      "loss": 6.4194,
+      "step": 6
+    },
+    {
+      "epoch": 0.05973333333333333,
+      "grad_norm": 71.37977490595638,
+      "learning_rate": 8.450980400142568e-06,
+      "loss": 5.4953,
+      "step": 7
+    },
+    {
+      "epoch": 0.06826666666666667,
+      "grad_norm": 49.31153456754566,
+      "learning_rate": 9.030899869919434e-06,
+      "loss": 5.4123,
+      "step": 8
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 20.37296364560341,
+      "learning_rate": 9.542425094393249e-06,
+      "loss": 5.2334,
+      "step": 9
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 7.968467079076881,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 5.0282,
+      "step": 10
+    },
+    {
+      "epoch": 0.09386666666666667,
+      "grad_norm": 3.559446532055649,
+      "learning_rate": 1.041392685158225e-05,
+      "loss": 4.612,
+      "step": 11
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 3.5528846947995674,
+      "learning_rate": 1.0791812460476248e-05,
+      "loss": 4.9475,
+      "step": 12
+    },
+    {
+      "epoch": 0.11093333333333333,
+      "grad_norm": 3.541968897471334,
+      "learning_rate": 1.1139433523068365e-05,
+      "loss": 4.2777,
+      "step": 13
+    },
+    {
+      "epoch": 0.11946666666666667,
+      "grad_norm": 3.54718070036198,
+      "learning_rate": 1.1461280356782378e-05,
+      "loss": 4.3507,
+      "step": 14
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 3.8632334830606747,
+      "learning_rate": 1.1760912590556813e-05,
+      "loss": 4.5364,
+      "step": 15
+    },
+    {
+      "epoch": 0.13653333333333334,
+      "grad_norm": 3.6637424744054004,
+      "learning_rate": 1.2041199826559246e-05,
+      "loss": 3.9672,
+      "step": 16
+    },
+    {
+      "epoch": 0.14506666666666668,
+      "grad_norm": 3.919802904818311,
+      "learning_rate": 1.230448921378274e-05,
+      "loss": 4.0618,
+      "step": 17
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 4.71904950738746,
+      "learning_rate": 1.2552725051033058e-05,
+      "loss": 4.6656,
+      "step": 18
+    },
+    {
+      "epoch": 0.16213333333333332,
+      "grad_norm": 4.6656317698690835,
+      "learning_rate": 1.2787536009528288e-05,
+      "loss": 4.1131,
+      "step": 19
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 5.145138692367417,
+      "learning_rate": 1.301029995663981e-05,
+      "loss": 4.0989,
+      "step": 20
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 5.923538014759818,
+      "learning_rate": 1.3222192947339192e-05,
+      "loss": 4.4991,
+      "step": 21
+    },
+    {
+      "epoch": 0.18773333333333334,
+      "grad_norm": 5.941056962941364,
+      "learning_rate": 1.3424226808222062e-05,
+      "loss": 4.0836,
+      "step": 22
+    },
+    {
+      "epoch": 0.19626666666666667,
+      "grad_norm": 6.171026012117947,
+      "learning_rate": 1.3617278360175927e-05,
+      "loss": 3.6861,
+      "step": 23
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 7.130542138930838,
+      "learning_rate": 1.380211241711606e-05,
+      "loss": 4.0958,
+      "step": 24
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 7.328837606110418,
+      "learning_rate": 1.3979400086720374e-05,
+      "loss": 3.9524,
+      "step": 25
+    },
+    {
+      "epoch": 0.22186666666666666,
+      "grad_norm": 6.923489005711429,
+      "learning_rate": 1.4149733479708178e-05,
+      "loss": 3.6062,
+      "step": 26
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 6.409498926059221,
+      "learning_rate": 1.4313637641589872e-05,
+      "loss": 3.2034,
+      "step": 27
+    },
+    {
+      "epoch": 0.23893333333333333,
+      "grad_norm": 5.781628405584682,
+      "learning_rate": 1.4471580313422191e-05,
+      "loss": 2.8158,
+      "step": 28
+    },
+    {
+      "epoch": 0.24746666666666667,
+      "grad_norm": 6.2927590068243315,
+      "learning_rate": 1.4623979978989559e-05,
+      "loss": 2.9803,
+      "step": 29
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 6.103844678630006,
+      "learning_rate": 1.4771212547196623e-05,
+      "loss": 2.847,
+      "step": 30
+    },
+    {
+      "epoch": 0.26453333333333334,
+      "grad_norm": 7.656341925867144,
+      "learning_rate": 1.4913616938342726e-05,
+      "loss": 3.0907,
+      "step": 31
+    },
+    {
+      "epoch": 0.2730666666666667,
+      "grad_norm": 6.324242877501844,
+      "learning_rate": 1.5051499783199059e-05,
+      "loss": 2.3467,
+      "step": 32
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 7.606313101162739,
+      "learning_rate": 1.5185139398778874e-05,
+      "loss": 2.5292,
+      "step": 33
+    },
+    {
+      "epoch": 0.29013333333333335,
+      "grad_norm": 8.553792493849265,
+      "learning_rate": 1.531478917042255e-05,
+      "loss": 2.4547,
+      "step": 34
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 8.483368703272543,
+      "learning_rate": 1.5440680443502753e-05,
+      "loss": 2.1956,
+      "step": 35
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 8.364739352838077,
+      "learning_rate": 1.5563025007672873e-05,
+      "loss": 1.8552,
+      "step": 36
+    },
+    {
+      "epoch": 0.3157333333333333,
+      "grad_norm": 9.37663682000104,
+      "learning_rate": 1.5682017240669948e-05,
+      "loss": 1.9228,
+      "step": 37
+    },
+    {
+      "epoch": 0.32426666666666665,
+      "grad_norm": 9.823047193440066,
+      "learning_rate": 1.57978359661681e-05,
+      "loss": 1.7033,
+      "step": 38
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 9.692618512955894,
+      "learning_rate": 1.591064607026499e-05,
+      "loss": 1.3768,
+      "step": 39
+    },
+    {
+      "epoch": 0.3413333333333333,
+      "grad_norm": 9.08889021911031,
+      "learning_rate": 1.6020599913279622e-05,
+      "loss": 1.3015,
+      "step": 40
+    },
+    {
+      "epoch": 0.34986666666666666,
+      "grad_norm": 8.081534221516058,
+      "learning_rate": 1.6127838567197353e-05,
+      "loss": 0.9228,
+      "step": 41
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 6.238638048950311,
+      "learning_rate": 1.6232492903979005e-05,
+      "loss": 0.7267,
+      "step": 42
+    },
+    {
+      "epoch": 0.36693333333333333,
+      "grad_norm": 3.4058036861773604,
+      "learning_rate": 1.6334684555795865e-05,
+      "loss": 0.5875,
+      "step": 43
+    },
+    {
+      "epoch": 0.37546666666666667,
+      "grad_norm": 2.079163829467713,
+      "learning_rate": 1.6434526764861872e-05,
+      "loss": 0.6355,
+      "step": 44
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 1.5597487833024746,
+      "learning_rate": 1.6532125137753435e-05,
+      "loss": 0.5106,
+      "step": 45
+    },
+    {
+      "epoch": 0.39253333333333335,
+      "grad_norm": 2.491689602375256,
+      "learning_rate": 1.662757831681574e-05,
+      "loss": 0.6454,
+      "step": 46
+    },
+    {
+      "epoch": 0.4010666666666667,
+      "grad_norm": 2.010880438195854,
+      "learning_rate": 1.672097857935717e-05,
+      "loss": 0.4757,
+      "step": 47
+    },
+    {
+      "epoch": 0.4096,
+      "grad_norm": 1.9452805114322096,
+      "learning_rate": 1.681241237375587e-05,
+      "loss": 0.4133,
+      "step": 48
+    },
+    {
+      "epoch": 0.41813333333333336,
+      "grad_norm": 1.7620848552306103,
+      "learning_rate": 1.6901960800285137e-05,
+      "loss": 0.4004,
+      "step": 49
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 1.278224489774809,
+      "learning_rate": 1.6989700043360187e-05,
+      "loss": 0.3523,
+      "step": 50
+    },
+    {
+      "epoch": 0.4352,
+      "grad_norm": 1.6151354758303231,
+      "learning_rate": 1.7075701760979363e-05,
+      "loss": 0.4317,
+      "step": 51
+    },
+    {
+      "epoch": 0.4437333333333333,
+      "grad_norm": 1.3451396055695035,
+      "learning_rate": 1.716003343634799e-05,
+      "loss": 0.3474,
+      "step": 52
+    },
+    {
+      "epoch": 0.45226666666666665,
+      "grad_norm": 1.6814977782362666,
+      "learning_rate": 1.724275869600789e-05,
+      "loss": 0.3706,
+      "step": 53
+    },
+    {
+      "epoch": 0.4608,
+      "grad_norm": 1.1682442432688667,
+      "learning_rate": 1.7323937598229687e-05,
+      "loss": 0.3488,
+      "step": 54
+    },
+    {
+      "epoch": 0.4693333333333333,
+      "grad_norm": 0.8839814540462471,
+      "learning_rate": 1.7403626894942437e-05,
+      "loss": 0.293,
+      "step": 55
+    },
+    {
+      "epoch": 0.47786666666666666,
+      "grad_norm": 0.7974304806295485,
+      "learning_rate": 1.7481880270062003e-05,
+      "loss": 0.2717,
+      "step": 56
+    },
+    {
+      "epoch": 0.4864,
+      "grad_norm": 0.6232203657585239,
+      "learning_rate": 1.7558748556724913e-05,
+      "loss": 0.1741,
+      "step": 57
+    },
+    {
+      "epoch": 0.49493333333333334,
+      "grad_norm": 0.6850688604563008,
+      "learning_rate": 1.763427993562937e-05,
+      "loss": 0.228,
+      "step": 58
+    },
+    {
+      "epoch": 0.5034666666666666,
+      "grad_norm": 0.5923826384300431,
+      "learning_rate": 1.7708520116421443e-05,
+      "loss": 0.2131,
+      "step": 59
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.489754430485032,
+      "learning_rate": 1.7781512503836432e-05,
+      "loss": 0.165,
+      "step": 60
+    },
+    {
+      "epoch": 0.5205333333333333,
+      "grad_norm": 0.5280940052395061,
+      "learning_rate": 1.7853298350107667e-05,
+      "loss": 0.1658,
+      "step": 61
+    },
+    {
+      "epoch": 0.5290666666666667,
+      "grad_norm": 0.4750905992036739,
+      "learning_rate": 1.7923916894982537e-05,
+      "loss": 0.1438,
+      "step": 62
+    },
+    {
+      "epoch": 0.5376,
+      "grad_norm": 0.4162998392722401,
+      "learning_rate": 1.7993405494535815e-05,
+      "loss": 0.1555,
+      "step": 63
+    },
+    {
+      "epoch": 0.5461333333333334,
+      "grad_norm": 0.26863266246370443,
+      "learning_rate": 1.806179973983887e-05,
+      "loss": 0.1323,
+      "step": 64
+    },
+    {
+      "epoch": 0.5546666666666666,
+      "grad_norm": 0.26534221125601215,
+      "learning_rate": 1.8129133566428553e-05,
+      "loss": 0.1671,
+      "step": 65
+    },
+    {
+      "epoch": 0.5632,
+      "grad_norm": 0.2548662962257576,
+      "learning_rate": 1.8195439355418686e-05,
+      "loss": 0.1308,
+      "step": 66
+    },
+    {
+      "epoch": 0.5717333333333333,
+      "grad_norm": 0.18045394638578796,
+      "learning_rate": 1.8260748027008263e-05,
+      "loss": 0.1262,
+      "step": 67
+    },
+    {
+      "epoch": 0.5802666666666667,
+      "grad_norm": 0.17070388073612064,
+      "learning_rate": 1.8325089127062364e-05,
+      "loss": 0.1192,
+      "step": 68
+    },
+    {
+      "epoch": 0.5888,
+      "grad_norm": 0.1531381679244776,
+      "learning_rate": 1.8388490907372553e-05,
+      "loss": 0.1274,
+      "step": 69
+    },
+    {
+      "epoch": 0.5973333333333334,
+      "grad_norm": 0.18196434993687946,
+      "learning_rate": 1.8450980400142568e-05,
+      "loss": 0.1375,
+      "step": 70
+    },
+    {
+      "epoch": 0.6058666666666667,
+      "grad_norm": 0.15324416972951205,
+      "learning_rate": 1.8512583487190752e-05,
+      "loss": 0.1599,
+      "step": 71
+    },
+    {
+      "epoch": 0.6144,
+      "grad_norm": 0.10884462064503801,
+      "learning_rate": 1.857332496431268e-05,
+      "loss": 0.1041,
+      "step": 72
+    },
+    {
+      "epoch": 0.6229333333333333,
+      "grad_norm": 0.12915133528192668,
+      "learning_rate": 1.8633228601204555e-05,
+      "loss": 0.1406,
+      "step": 73
+    },
+    {
+      "epoch": 0.6314666666666666,
+      "grad_norm": 0.12553425699952878,
+      "learning_rate": 1.8692317197309763e-05,
+      "loss": 0.1256,
+      "step": 74
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11976960918968543,
+      "learning_rate": 1.8750612633916997e-05,
+      "loss": 0.1144,
+      "step": 75
+    },
+    {
+      "epoch": 0.6485333333333333,
+      "grad_norm": 0.115805998298789,
+      "learning_rate": 1.8808135922807914e-05,
+      "loss": 0.1528,
+      "step": 76
+    },
+    {
+      "epoch": 0.6570666666666667,
+      "grad_norm": 0.10325948496697443,
+      "learning_rate": 1.8864907251724818e-05,
+      "loss": 0.1044,
+      "step": 77
+    },
+    {
+      "epoch": 0.6656,
+      "grad_norm": 0.09595064346541006,
+      "learning_rate": 1.8920946026904802e-05,
+      "loss": 0.1534,
+      "step": 78
+    },
+    {
+      "epoch": 0.6741333333333334,
+      "grad_norm": 0.08796742845240496,
+      "learning_rate": 1.8976270912904414e-05,
+      "loss": 0.1155,
+      "step": 79
+    },
+    {
+      "epoch": 0.6826666666666666,
+      "grad_norm": 0.08218991738379527,
+      "learning_rate": 1.9030899869919434e-05,
+      "loss": 0.1311,
+      "step": 80
+    },
+    {
+      "epoch": 0.6912,
+      "grad_norm": 0.08290815261109215,
+      "learning_rate": 1.9084850188786497e-05,
+      "loss": 0.11,
+      "step": 81
+    },
+    {
+      "epoch": 0.6997333333333333,
+      "grad_norm": 0.0794132180176064,
+      "learning_rate": 1.9138138523837165e-05,
+      "loss": 0.1135,
+      "step": 82
+    },
+    {
+      "epoch": 0.7082666666666667,
+      "grad_norm": 0.06934410705255296,
+      "learning_rate": 1.919078092376074e-05,
+      "loss": 0.109,
+      "step": 83
+    },
+    {
+      "epoch": 0.7168,
+      "grad_norm": 0.09000563031870593,
+      "learning_rate": 1.9242792860618813e-05,
+      "loss": 0.12,
+      "step": 84
+    },
+    {
+      "epoch": 0.7253333333333334,
+      "grad_norm": 0.1134042277879818,
+      "learning_rate": 1.929418925714293e-05,
+      "loss": 0.1223,
+      "step": 85
+    },
+    {
+      "epoch": 0.7338666666666667,
+      "grad_norm": 0.09118764690233076,
+      "learning_rate": 1.9344984512435673e-05,
+      "loss": 0.1459,
+      "step": 86
+    },
+    {
+      "epoch": 0.7424,
+      "grad_norm": 0.07873016754353963,
+      "learning_rate": 1.9395192526186183e-05,
+      "loss": 0.1422,
+      "step": 87
+    },
+    {
+      "epoch": 0.7509333333333333,
+      "grad_norm": 0.1796495874463076,
+      "learning_rate": 1.9444826721501687e-05,
+      "loss": 0.1291,
+      "step": 88
+    },
+    {
+      "epoch": 0.7594666666666666,
+      "grad_norm": 0.0679589944174269,
+      "learning_rate": 1.9493900066449125e-05,
+      "loss": 0.108,
+      "step": 89
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.08174688574235538,
+      "learning_rate": 1.9542425094393246e-05,
+      "loss": 0.1081,
+      "step": 90
+    },
+    {
+      "epoch": 0.7765333333333333,
+      "grad_norm": 0.057137370501406756,
+      "learning_rate": 1.9590413923210934e-05,
+      "loss": 0.0934,
+      "step": 91
+    },
+    {
+      "epoch": 0.7850666666666667,
+      "grad_norm": 0.06578111924908255,
+      "learning_rate": 1.9637878273455555e-05,
+      "loss": 0.1085,
+      "step": 92
+    },
+    {
+      "epoch": 0.7936,
+      "grad_norm": 0.08945990540906254,
+      "learning_rate": 1.968482948553935e-05,
+      "loss": 0.1747,
+      "step": 93
+    },
+    {
+      "epoch": 0.8021333333333334,
+      "grad_norm": 0.06183863311044229,
+      "learning_rate": 1.9731278535996986e-05,
+      "loss": 0.1136,
+      "step": 94
+    },
+    {
+      "epoch": 0.8106666666666666,
+      "grad_norm": 0.05777899602544702,
+      "learning_rate": 1.9777236052888476e-05,
+      "loss": 0.0984,
+      "step": 95
+    },
+    {
+      "epoch": 0.8192,
+      "grad_norm": 0.08130851607693534,
+      "learning_rate": 1.9822712330395683e-05,
+      "loss": 0.187,
+      "step": 96
+    },
+    {
+      "epoch": 0.8277333333333333,
+      "grad_norm": 0.06426546202002927,
+      "learning_rate": 1.986771734266245e-05,
+      "loss": 0.1296,
+      "step": 97
+    },
+    {
+      "epoch": 0.8362666666666667,
+      "grad_norm": 0.069692313707994,
+      "learning_rate": 1.991226075692495e-05,
+      "loss": 0.1404,
+      "step": 98
+    },
+    {
+      "epoch": 0.8448,
+      "grad_norm": 0.05494542266886729,
+      "learning_rate": 1.9956351945975496e-05,
+      "loss": 0.116,
+      "step": 99
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.07571686966840627,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 0.1539,
+      "step": 100
+    },
+    {
+      "epoch": 0.8618666666666667,
+      "grad_norm": 0.054351059117603705,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 101
+    },
+    {
+      "epoch": 0.8704,
+      "grad_norm": 0.06531899611551092,
+      "learning_rate": 2e-05,
+      "loss": 0.0827,
+      "step": 102
+    },
+    {
+      "epoch": 0.8789333333333333,
+      "grad_norm": 0.06131678646504652,
+      "learning_rate": 2e-05,
+      "loss": 0.1266,
+      "step": 103
+    },
+    {
+      "epoch": 0.8874666666666666,
+      "grad_norm": 0.06850220540661824,
+      "learning_rate": 2e-05,
+      "loss": 0.1456,
+      "step": 104
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.05806908951252483,
+      "learning_rate": 2e-05,
+      "loss": 0.0954,
+      "step": 105
+    },
+    {
+      "epoch": 0.9045333333333333,
+      "grad_norm": 0.06503642452033717,
+      "learning_rate": 2e-05,
+      "loss": 0.1417,
+      "step": 106
+    },
+    {
+      "epoch": 0.9130666666666667,
+      "grad_norm": 0.050486271853277066,
+      "learning_rate": 2e-05,
+      "loss": 0.0959,
+      "step": 107
+    },
+    {
+      "epoch": 0.9216,
+      "grad_norm": 0.07746063813802379,
+      "learning_rate": 2e-05,
+      "loss": 0.1256,
+      "step": 108
+    },
+    {
+      "epoch": 0.9301333333333334,
+      "grad_norm": 0.051231172380840004,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 109
+    },
+    {
+      "epoch": 0.9386666666666666,
+      "grad_norm": 0.056296443557859455,
+      "learning_rate": 2e-05,
+      "loss": 0.1056,
+      "step": 110
+    },
+    {
+      "epoch": 0.9472,
+      "grad_norm": 0.05058663240713958,
+      "learning_rate": 2e-05,
+      "loss": 0.0971,
+      "step": 111
+    },
+    {
+      "epoch": 0.9557333333333333,
+      "grad_norm": 0.05532886570130611,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 112
+    },
+    {
+      "epoch": 0.9642666666666667,
+      "grad_norm": 0.05327811326654907,
+      "learning_rate": 2e-05,
+      "loss": 0.0989,
+      "step": 113
+    },
+    {
+      "epoch": 0.9728,
+      "grad_norm": 0.05663279364147864,
+      "learning_rate": 2e-05,
+      "loss": 0.0958,
+      "step": 114
+    },
+    {
+      "epoch": 0.9813333333333333,
+      "grad_norm": 0.04930904541225805,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 115
+    },
+    {
+      "epoch": 0.9898666666666667,
+      "grad_norm": 0.06096947951115022,
+      "learning_rate": 2e-05,
+      "loss": 0.106,
+      "step": 116
+    },
+    {
+      "epoch": 0.9984,
+      "grad_norm": 0.050092322361182495,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 117
+    },
+    {
+      "epoch": 1.0069333333333332,
+      "grad_norm": 0.04980408443758999,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 118
+    },
+    {
+      "epoch": 1.0154666666666667,
+      "grad_norm": 0.051183082721834305,
+      "learning_rate": 2e-05,
+      "loss": 0.1049,
+      "step": 119
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.04332220265802814,
+      "learning_rate": 2e-05,
+      "loss": 0.0983,
+      "step": 120
+    },
+    {
+      "epoch": 1.0325333333333333,
+      "grad_norm": 0.07211731499677299,
+      "learning_rate": 2e-05,
+      "loss": 0.1386,
+      "step": 121
+    },
+    {
+      "epoch": 1.0410666666666666,
+      "grad_norm": 0.06550870223740553,
+      "learning_rate": 2e-05,
+      "loss": 0.1334,
+      "step": 122
+    },
+    {
+      "epoch": 1.0496,
+      "grad_norm": 0.05331839690767287,
+      "learning_rate": 2e-05,
+      "loss": 0.1014,
+      "step": 123
+    },
+    {
+      "epoch": 1.0581333333333334,
+      "grad_norm": 0.05227685628767905,
+      "learning_rate": 2e-05,
+      "loss": 0.1098,
+      "step": 124
+    },
+    {
+      "epoch": 1.0666666666666667,
+      "grad_norm": 0.07641693882491171,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 125
+    },
+    {
+      "epoch": 1.0752,
+      "grad_norm": 0.052835367770791786,
+      "learning_rate": 2e-05,
+      "loss": 0.1231,
+      "step": 126
+    },
+    {
+      "epoch": 1.0837333333333334,
+      "grad_norm": 0.07520329755025788,
+      "learning_rate": 2e-05,
+      "loss": 0.085,
+      "step": 127
+    },
+    {
+      "epoch": 1.0922666666666667,
+      "grad_norm": 0.07670066152157425,
+      "learning_rate": 2e-05,
+      "loss": 0.1071,
+      "step": 128
+    },
+    {
+      "epoch": 1.1008,
+      "grad_norm": 0.052832906560645154,
+      "learning_rate": 2e-05,
+      "loss": 0.1093,
+      "step": 129
+    },
+    {
+      "epoch": 1.1093333333333333,
+      "grad_norm": 0.06573889037311398,
+      "learning_rate": 2e-05,
+      "loss": 0.1193,
+      "step": 130
+    },
+    {
+      "epoch": 1.1178666666666666,
+      "grad_norm": 0.05175471296566334,
+      "learning_rate": 2e-05,
+      "loss": 0.1184,
+      "step": 131
+    },
+    {
+      "epoch": 1.1264,
+      "grad_norm": 0.05912231419793496,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 132
+    },
+    {
+      "epoch": 1.1349333333333333,
+      "grad_norm": 0.04899140475981105,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 133
+    },
+    {
+      "epoch": 1.1434666666666666,
+      "grad_norm": 0.05939916939142137,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 134
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.0516819217599706,
+      "learning_rate": 2e-05,
+      "loss": 0.0834,
+      "step": 135
+    },
+    {
+      "epoch": 1.1605333333333334,
+      "grad_norm": 0.05456440346454737,
+      "learning_rate": 2e-05,
+      "loss": 0.1183,
+      "step": 136
+    },
+    {
+      "epoch": 1.1690666666666667,
+      "grad_norm": 0.059906464476343235,
+      "learning_rate": 2e-05,
+      "loss": 0.1048,
+      "step": 137
+    },
+    {
+      "epoch": 1.1776,
+      "grad_norm": 0.0720112680204319,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 138
+    },
+    {
+      "epoch": 1.1861333333333333,
+      "grad_norm": 0.04940202805828527,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 139
+    },
+    {
+      "epoch": 1.1946666666666665,
+      "grad_norm": 0.060088609545130046,
+      "learning_rate": 2e-05,
+      "loss": 0.0952,
+      "step": 140
+    },
+    {
+      "epoch": 1.2032,
+      "grad_norm": 0.04694761423612446,
+      "learning_rate": 2e-05,
+      "loss": 0.0717,
+      "step": 141
+    },
+    {
+      "epoch": 1.2117333333333333,
+      "grad_norm": 0.05628581562512457,
+      "learning_rate": 2e-05,
+      "loss": 0.1062,
+      "step": 142
+    },
+    {
+      "epoch": 1.2202666666666666,
+      "grad_norm": 0.06876420990437652,
+      "learning_rate": 2e-05,
+      "loss": 0.1218,
+      "step": 143
+    },
+    {
+      "epoch": 1.2288000000000001,
+      "grad_norm": 0.058774700501610655,
+      "learning_rate": 2e-05,
+      "loss": 0.1125,
+      "step": 144
+    },
+    {
+      "epoch": 1.2373333333333334,
+      "grad_norm": 0.061855922064341186,
+      "learning_rate": 2e-05,
+      "loss": 0.1295,
+      "step": 145
+    },
+    {
+      "epoch": 1.2458666666666667,
+      "grad_norm": 0.0813047704730138,
+      "learning_rate": 2e-05,
+      "loss": 0.1165,
+      "step": 146
+    },
+    {
+      "epoch": 1.2544,
+      "grad_norm": 0.061374000305305446,
+      "learning_rate": 2e-05,
+      "loss": 0.1094,
+      "step": 147
+    },
+    {
+      "epoch": 1.2629333333333332,
+      "grad_norm": 0.055537169110833,
+      "learning_rate": 2e-05,
+      "loss": 0.1054,
+      "step": 148
+    },
+    {
+      "epoch": 1.2714666666666667,
+      "grad_norm": 0.04423248714119304,
+      "learning_rate": 2e-05,
+      "loss": 0.0841,
+      "step": 149
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.049931966607835034,
+      "learning_rate": 2e-05,
+      "loss": 0.0961,
+      "step": 150
+    },
+    {
+      "epoch": 1.2885333333333333,
+      "grad_norm": 0.06178656953298769,
+      "learning_rate": 2e-05,
+      "loss": 0.0854,
+      "step": 151
+    },
+    {
+      "epoch": 1.2970666666666666,
+      "grad_norm": 0.05783812343287897,
+      "learning_rate": 2e-05,
+      "loss": 0.1141,
+      "step": 152
+    },
+    {
+      "epoch": 1.3056,
+      "grad_norm": 0.048955120400167584,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 153
+    },
+    {
+      "epoch": 1.3141333333333334,
+      "grad_norm": 0.12270174746806978,
+      "learning_rate": 2e-05,
+      "loss": 0.1553,
+      "step": 154
+    },
+    {
+      "epoch": 1.3226666666666667,
+      "grad_norm": 0.06928026959973474,
+      "learning_rate": 2e-05,
+      "loss": 0.1274,
+      "step": 155
+    },
+    {
+      "epoch": 1.3312,
+      "grad_norm": 0.04756100666105405,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 156
+    },
+    {
+      "epoch": 1.3397333333333332,
+      "grad_norm": 0.056054951338196934,
+      "learning_rate": 2e-05,
+      "loss": 0.0831,
+      "step": 157
+    },
+    {
+      "epoch": 1.3482666666666667,
+      "grad_norm": 0.0516990471964239,
+      "learning_rate": 2e-05,
+      "loss": 0.0883,
+      "step": 158
+    },
+    {
+      "epoch": 1.3568,
+      "grad_norm": 0.06011650542069954,
+      "learning_rate": 2e-05,
+      "loss": 0.0938,
+      "step": 159
+    },
+    {
+      "epoch": 1.3653333333333333,
+      "grad_norm": 0.051831307951873976,
+      "learning_rate": 2e-05,
+      "loss": 0.1019,
+      "step": 160
+    },
+    {
+      "epoch": 1.3738666666666668,
+      "grad_norm": 0.0544902702048936,
+      "learning_rate": 2e-05,
+      "loss": 0.0906,
+      "step": 161
+    },
+    {
+      "epoch": 1.3824,
+      "grad_norm": 0.06057617032526,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 162
+    },
+    {
+      "epoch": 1.3909333333333334,
+      "grad_norm": 0.07288058025189605,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 163
+    },
+    {
+      "epoch": 1.3994666666666666,
+      "grad_norm": 0.05639043792084219,
+      "learning_rate": 2e-05,
+      "loss": 0.1031,
+      "step": 164
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.0586469408837505,
+      "learning_rate": 2e-05,
+      "loss": 0.1143,
+      "step": 165
+    },
+    {
+      "epoch": 1.4165333333333332,
+      "grad_norm": 0.059243429040783904,
+      "learning_rate": 2e-05,
+      "loss": 0.0838,
+      "step": 166
+    },
+    {
+      "epoch": 1.4250666666666667,
+      "grad_norm": 0.0621476448363388,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 167
+    },
+    {
+      "epoch": 1.4336,
+      "grad_norm": 0.06083867682720169,
+      "learning_rate": 2e-05,
+      "loss": 0.1119,
+      "step": 168
+    },
+    {
+      "epoch": 1.4421333333333333,
+      "grad_norm": 0.09583165335305677,
+      "learning_rate": 2e-05,
+      "loss": 0.1028,
+      "step": 169
+    },
+    {
+      "epoch": 1.4506666666666668,
+      "grad_norm": 0.06411638581314043,
+      "learning_rate": 2e-05,
+      "loss": 0.1181,
+      "step": 170
+    },
+    {
+      "epoch": 1.4592,
+      "grad_norm": 0.05632977234908945,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 171
+    },
+    {
+      "epoch": 1.4677333333333333,
+      "grad_norm": 0.05666068971337413,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 172
+    },
+    {
+      "epoch": 1.4762666666666666,
+      "grad_norm": 0.04502062723807536,
+      "learning_rate": 2e-05,
+      "loss": 0.0588,
+      "step": 173
+    },
+    {
+      "epoch": 1.4848,
+      "grad_norm": 0.05916500176868301,
+      "learning_rate": 2e-05,
+      "loss": 0.0949,
+      "step": 174
+    },
+    {
+      "epoch": 1.4933333333333334,
+      "grad_norm": 0.056484273808864845,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 175
+    },
+    {
+      "epoch": 1.5018666666666667,
+      "grad_norm": 0.06652084448571674,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 176
+    },
+    {
+      "epoch": 1.5104,
+      "grad_norm": 0.05400238078068516,
+      "learning_rate": 2e-05,
+      "loss": 0.0919,
+      "step": 177
+    },
+    {
+      "epoch": 1.5189333333333335,
+      "grad_norm": 0.04979579743346662,
+      "learning_rate": 2e-05,
+      "loss": 0.0879,
+      "step": 178
+    },
+    {
+      "epoch": 1.5274666666666668,
+      "grad_norm": 0.06876105414733971,
+      "learning_rate": 2e-05,
+      "loss": 0.1162,
+      "step": 179
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.05633262015499721,
+      "learning_rate": 2e-05,
+      "loss": 0.1142,
+      "step": 180
+    },
+    {
+      "epoch": 1.5445333333333333,
+      "grad_norm": 0.0599508967519892,
+      "learning_rate": 2e-05,
+      "loss": 0.1073,
+      "step": 181
+    },
+    {
+      "epoch": 1.5530666666666666,
+      "grad_norm": 0.058912170976454126,
+      "learning_rate": 2e-05,
+      "loss": 0.1102,
+      "step": 182
+    },
+    {
+      "epoch": 1.5615999999999999,
+      "grad_norm": 0.05361414988566254,
+      "learning_rate": 2e-05,
+      "loss": 0.0885,
+      "step": 183
+    },
+    {
+      "epoch": 1.5701333333333334,
+      "grad_norm": 0.04517277847384077,
+      "learning_rate": 2e-05,
+      "loss": 0.0763,
+      "step": 184
+    },
+    {
+      "epoch": 1.5786666666666667,
+      "grad_norm": 0.05551553662051771,
+      "learning_rate": 2e-05,
+      "loss": 0.0877,
+      "step": 185
+    },
+    {
+      "epoch": 1.5872000000000002,
+      "grad_norm": 0.05814223969236194,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 186
+    },
+    {
+      "epoch": 1.5957333333333334,
+      "grad_norm": 0.05710054166191597,
+      "learning_rate": 2e-05,
+      "loss": 0.0962,
+      "step": 187
+    },
+    {
+      "epoch": 1.6042666666666667,
+      "grad_norm": 0.054744343932104075,
+      "learning_rate": 2e-05,
+      "loss": 0.0873,
+      "step": 188
+    },
+    {
+      "epoch": 1.6128,
+      "grad_norm": 0.051145521687090995,
+      "learning_rate": 2e-05,
+      "loss": 0.0855,
+      "step": 189
+    },
+    {
+      "epoch": 1.6213333333333333,
+      "grad_norm": 0.05414658860737789,
+      "learning_rate": 2e-05,
+      "loss": 0.0872,
+      "step": 190
+    },
+    {
+      "epoch": 1.6298666666666666,
+      "grad_norm": 0.05084744595533029,
+      "learning_rate": 2e-05,
+      "loss": 0.0891,
+      "step": 191
+    },
+    {
+      "epoch": 1.6383999999999999,
+      "grad_norm": 0.0567070082820597,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 192
+    },
+    {
+      "epoch": 1.6469333333333334,
+      "grad_norm": 0.0494785311411315,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 193
+    },
+    {
+      "epoch": 1.6554666666666666,
+      "grad_norm": 0.062341158530385396,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 194
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.059888336716275685,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 195
+    },
+    {
+      "epoch": 1.6725333333333334,
+      "grad_norm": 0.07346562318829057,
+      "learning_rate": 2e-05,
+      "loss": 0.1329,
+      "step": 196
+    },
+    {
+      "epoch": 1.6810666666666667,
+      "grad_norm": 0.0792360016934733,
+      "learning_rate": 2e-05,
+      "loss": 0.1392,
+      "step": 197
+    },
+    {
+      "epoch": 1.6896,
+      "grad_norm": 0.0800342963229883,
+      "learning_rate": 2e-05,
+      "loss": 0.1199,
+      "step": 198
+    },
+    {
+      "epoch": 1.6981333333333333,
+      "grad_norm": 0.06848045933195548,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 199
+    },
+    {
+      "epoch": 1.7066666666666666,
+      "grad_norm": 0.05743199019316764,
+      "learning_rate": 2e-05,
+      "loss": 0.0811,
+      "step": 200
+    },
+    {
+      "epoch": 1.7151999999999998,
+      "grad_norm": 0.07170531168284446,
+      "learning_rate": 2e-05,
+      "loss": 0.1079,
+      "step": 201
+    },
+    {
+      "epoch": 1.7237333333333333,
+      "grad_norm": 0.05772905481368506,
+      "learning_rate": 2e-05,
+      "loss": 0.0844,
+      "step": 202
+    },
+    {
+      "epoch": 1.7322666666666666,
+      "grad_norm": 0.07504946014098464,
+      "learning_rate": 2e-05,
+      "loss": 0.1257,
+      "step": 203
+    },
+    {
+      "epoch": 1.7408000000000001,
+      "grad_norm": 0.06450179819785838,
+      "learning_rate": 2e-05,
+      "loss": 0.1104,
+      "step": 204
+    },
+    {
+      "epoch": 1.7493333333333334,
+      "grad_norm": 0.06595445965110332,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 205
+    },
+    {
+      "epoch": 1.7578666666666667,
+      "grad_norm": 0.07203558121131749,
+      "learning_rate": 2e-05,
+      "loss": 0.1117,
+      "step": 206
+    },
+    {
+      "epoch": 1.7664,
+      "grad_norm": 0.05954646782409283,
+      "learning_rate": 2e-05,
+      "loss": 0.0729,
+      "step": 207
+    },
+    {
+      "epoch": 1.7749333333333333,
+      "grad_norm": 0.06624894584410884,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 208
+    },
+    {
+      "epoch": 1.7834666666666665,
+      "grad_norm": 0.06888562028256219,
+      "learning_rate": 2e-05,
+      "loss": 0.1398,
+      "step": 209
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.061224088077794406,
+      "learning_rate": 2e-05,
+      "loss": 0.1112,
+      "step": 210
+    },
+    {
+      "epoch": 1.8005333333333333,
+      "grad_norm": 0.06857358883856608,
+      "learning_rate": 2e-05,
+      "loss": 0.1293,
+      "step": 211
+    },
+    {
+      "epoch": 1.8090666666666668,
+      "grad_norm": 0.06177352416779139,
+      "learning_rate": 2e-05,
+      "loss": 0.0884,
+      "step": 212
+    },
+    {
+      "epoch": 1.8176,
+      "grad_norm": 0.08324567429925228,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 213
+    },
+    {
+      "epoch": 1.8261333333333334,
+      "grad_norm": 0.06771677297787752,
+      "learning_rate": 2e-05,
+      "loss": 0.089,
+      "step": 214
+    },
+    {
+      "epoch": 1.8346666666666667,
+      "grad_norm": 0.07055754809472485,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 215
+    },
+    {
+      "epoch": 1.8432,
+      "grad_norm": 0.05856797724392531,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 216
+    },
+    {
+      "epoch": 1.8517333333333332,
+      "grad_norm": 0.07555286129801597,
+      "learning_rate": 2e-05,
+      "loss": 0.0913,
+      "step": 217
+    },
+    {
+      "epoch": 1.8602666666666665,
+      "grad_norm": 0.09242462538643775,
+      "learning_rate": 2e-05,
+      "loss": 0.1241,
+      "step": 218
+    },
+    {
+      "epoch": 1.8688,
+      "grad_norm": 0.06550805537088374,
+      "learning_rate": 2e-05,
+      "loss": 0.0819,
+      "step": 219
+    },
+    {
+      "epoch": 1.8773333333333333,
+      "grad_norm": 0.06016048263236861,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 220
+    },
+    {
+      "epoch": 1.8858666666666668,
+      "grad_norm": 0.06856661106001445,
+      "learning_rate": 2e-05,
+      "loss": 0.1132,
+      "step": 221
+    },
+    {
+      "epoch": 1.8944,
+      "grad_norm": 0.06383306455000529,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 222
+    },
+    {
+      "epoch": 1.9029333333333334,
+      "grad_norm": 0.07240472757239624,
+      "learning_rate": 2e-05,
+      "loss": 0.0863,
+      "step": 223
+    },
+    {
+      "epoch": 1.9114666666666666,
+      "grad_norm": 0.07845654471077741,
+      "learning_rate": 2e-05,
+      "loss": 0.1284,
+      "step": 224
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.07192185833649212,
+      "learning_rate": 2e-05,
+      "loss": 0.1101,
+      "step": 225
+    },
+    {
+      "epoch": 1.9285333333333332,
+      "grad_norm": 0.06390598005596872,
+      "learning_rate": 2e-05,
+      "loss": 0.0917,
+      "step": 226
+    },
+    {
+      "epoch": 1.9370666666666667,
+      "grad_norm": 0.06306138712432224,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 227
+    },
+    {
+      "epoch": 1.9456,
+      "grad_norm": 0.06771381941296478,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 228
+    },
+    {
+      "epoch": 1.9541333333333335,
+      "grad_norm": 0.05899006803461524,
+      "learning_rate": 2e-05,
+      "loss": 0.0782,
+      "step": 229
+    },
+    {
+      "epoch": 1.9626666666666668,
+      "grad_norm": 0.07426956281950735,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 230
+    },
+    {
+      "epoch": 1.9712,
+      "grad_norm": 0.06567534902293475,
+      "learning_rate": 2e-05,
+      "loss": 0.0956,
+      "step": 231
+    },
+    {
+      "epoch": 1.9797333333333333,
+      "grad_norm": 0.07430395142282198,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 232
+    },
+    {
+      "epoch": 1.9882666666666666,
+      "grad_norm": 0.05834447367264806,
+      "learning_rate": 2e-05,
+      "loss": 0.0767,
+      "step": 233
+    },
+    {
+      "epoch": 1.9968,
+      "grad_norm": 0.07137090413877054,
+      "learning_rate": 2e-05,
+      "loss": 0.0821,
+      "step": 234
+    },
+    {
+      "epoch": 2.005333333333333,
+      "grad_norm": 0.07797914240805551,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 235
+    },
+    {
+      "epoch": 2.0138666666666665,
+      "grad_norm": 0.09335648879374187,
+      "learning_rate": 2e-05,
+      "loss": 0.1632,
+      "step": 236
+    },
+    {
+      "epoch": 2.0224,
+      "grad_norm": 0.08280719749100944,
+      "learning_rate": 2e-05,
+      "loss": 0.115,
+      "step": 237
+    },
+    {
+      "epoch": 2.0309333333333335,
+      "grad_norm": 0.08956213539053312,
+      "learning_rate": 2e-05,
+      "loss": 0.109,
+      "step": 238
+    },
+    {
+      "epoch": 2.0394666666666668,
+      "grad_norm": 0.08849786687188893,
+      "learning_rate": 2e-05,
+      "loss": 0.1234,
+      "step": 239
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.16956065138312856,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 240
+    },
+    {
+      "epoch": 2.0565333333333333,
+      "grad_norm": 0.07377723133015779,
+      "learning_rate": 2e-05,
+      "loss": 0.1068,
+      "step": 241
+    },
+    {
+      "epoch": 2.0650666666666666,
+      "grad_norm": 0.08839651322303756,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 242
+    },
+    {
+      "epoch": 2.0736,
+      "grad_norm": 0.10641135540308218,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 243
+    },
+    {
+      "epoch": 2.082133333333333,
+      "grad_norm": 0.09370442613891326,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 244
+    },
+    {
+      "epoch": 2.0906666666666665,
+      "grad_norm": 0.08988093042206098,
+      "learning_rate": 2e-05,
+      "loss": 0.1201,
+      "step": 245
+    },
+    {
+      "epoch": 2.0992,
+      "grad_norm": 0.07167309696967938,
+      "learning_rate": 2e-05,
+      "loss": 0.0747,
+      "step": 246
+    },
+    {
+      "epoch": 2.1077333333333335,
+      "grad_norm": 0.08822161286881192,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 247
+    },
+    {
+      "epoch": 2.1162666666666667,
+      "grad_norm": 0.08439713613775747,
+      "learning_rate": 2e-05,
+      "loss": 0.1063,
+      "step": 248
+    },
+    {
+      "epoch": 2.1248,
+      "grad_norm": 0.0942232139678121,
+      "learning_rate": 2e-05,
+      "loss": 0.0865,
+      "step": 249
+    },
+    {
+      "epoch": 2.1333333333333333,
+      "grad_norm": 0.1066306844906037,
+      "learning_rate": 2e-05,
+      "loss": 0.1198,
+      "step": 250
+    },
+    {
+      "epoch": 2.1418666666666666,
+      "grad_norm": 0.10435492421714423,
+      "learning_rate": 2e-05,
+      "loss": 0.0968,
+      "step": 251
+    },
+    {
+      "epoch": 2.1504,
+      "grad_norm": 0.10513719348492025,
+      "learning_rate": 2e-05,
+      "loss": 0.1045,
+      "step": 252
+    },
+    {
+      "epoch": 2.158933333333333,
+      "grad_norm": 0.12580299248756027,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 253
+    },
+    {
+      "epoch": 2.167466666666667,
+      "grad_norm": 0.10265943183351584,
+      "learning_rate": 2e-05,
+      "loss": 0.0892,
+      "step": 254
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.09473119803029131,
+      "learning_rate": 2e-05,
+      "loss": 0.0721,
+      "step": 255
+    },
+    {
+      "epoch": 2.1845333333333334,
+      "grad_norm": 0.13420988295622846,
+      "learning_rate": 2e-05,
+      "loss": 0.1034,
+      "step": 256
+    },
+    {
+      "epoch": 2.1930666666666667,
+      "grad_norm": 0.11474592330595303,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 257
+    },
+    {
+      "epoch": 2.2016,
+      "grad_norm": 0.12382914500613644,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 258
+    },
+    {
+      "epoch": 2.2101333333333333,
+      "grad_norm": 0.12984065755657193,
+      "learning_rate": 2e-05,
+      "loss": 0.0888,
+      "step": 259
+    },
+    {
+      "epoch": 2.2186666666666666,
+      "grad_norm": 0.14210964340218168,
+      "learning_rate": 2e-05,
+      "loss": 0.105,
+      "step": 260
+    },
+    {
+      "epoch": 2.2272,
+      "grad_norm": 0.13862185852250883,
+      "learning_rate": 2e-05,
+      "loss": 0.0876,
+      "step": 261
+    },
+    {
+      "epoch": 2.235733333333333,
+      "grad_norm": 0.17284047904822364,
+      "learning_rate": 2e-05,
+      "loss": 0.0935,
+      "step": 262
+    },
+    {
+      "epoch": 2.244266666666667,
+      "grad_norm": 0.1409465306152598,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 263
+    },
+    {
+      "epoch": 2.2528,
+      "grad_norm": 0.15551418258816738,
+      "learning_rate": 2e-05,
+      "loss": 0.0871,
+      "step": 264
+    },
+    {
+      "epoch": 2.2613333333333334,
+      "grad_norm": 0.1741617194945442,
+      "learning_rate": 2e-05,
+      "loss": 0.098,
+      "step": 265
+    },
+    {
+      "epoch": 2.2698666666666667,
+      "grad_norm": 0.18387218423978585,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 266
+    },
+    {
+      "epoch": 2.2784,
+      "grad_norm": 0.21744684782545962,
+      "learning_rate": 2e-05,
+      "loss": 0.1061,
+      "step": 267
+    },
+    {
+      "epoch": 2.2869333333333333,
+      "grad_norm": 0.17372673806896413,
+      "learning_rate": 2e-05,
+      "loss": 0.0927,
+      "step": 268
+    },
+    {
+      "epoch": 2.2954666666666665,
+      "grad_norm": 0.20890400134226653,
+      "learning_rate": 2e-05,
+      "loss": 0.0874,
+      "step": 269
+    },
+    {
+      "epoch": 2.304,
+      "grad_norm": 0.20279172920289834,
+      "learning_rate": 2e-05,
+      "loss": 0.0807,
+      "step": 270
+    },
+    {
+      "epoch": 2.3125333333333336,
+      "grad_norm": 0.23060543242200862,
+      "learning_rate": 2e-05,
+      "loss": 0.0937,
+      "step": 271
+    },
+    {
+      "epoch": 2.321066666666667,
+      "grad_norm": 0.2866337268131343,
+      "learning_rate": 2e-05,
+      "loss": 0.0886,
+      "step": 272
+    },
+    {
+      "epoch": 2.3296,
+      "grad_norm": 0.3203014162518504,
+      "learning_rate": 2e-05,
+      "loss": 0.0741,
+      "step": 273
+    },
+    {
+      "epoch": 2.3381333333333334,
+      "grad_norm": 0.2702455454246614,
+      "learning_rate": 2e-05,
+      "loss": 0.0692,
+      "step": 274
+    },
+    {
+      "epoch": 2.3466666666666667,
+      "grad_norm": 0.30682595841991545,
+      "learning_rate": 2e-05,
+      "loss": 0.0733,
+      "step": 275
+    },
+    {
+      "epoch": 2.3552,
+      "grad_norm": 0.2844580109974713,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 276
+    },
+    {
+      "epoch": 2.3637333333333332,
+      "grad_norm": 0.31296149985818705,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 277
+    },
+    {
+      "epoch": 2.3722666666666665,
+      "grad_norm": 0.24448590086634314,
+      "learning_rate": 2e-05,
+      "loss": 0.1078,
+      "step": 278
+    },
+    {
+      "epoch": 2.3808,
+      "grad_norm": 0.17626685777153123,
+      "learning_rate": 2e-05,
+      "loss": 0.0939,
+      "step": 279
+    },
+    {
+      "epoch": 2.389333333333333,
+      "grad_norm": 0.13337416253457104,
+      "learning_rate": 2e-05,
+      "loss": 0.0812,
+      "step": 280
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 351,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.004836973419168e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-280/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa4edad1da5557fcd52a6da980443588016cead6f0444a3562cfa68029c66a04
+size 6840

checkpoint-280/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)

checkpoint-300/README.md

@@ -0,0 +1,202 @@
+---
+library_name: peft
+base_model: ../ckpts/Meta-Llama-3-8B-Instruct
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]
+### Framework versions
+
+- PEFT 0.11.1

checkpoint-300/adapter_config.json

@@ -0,0 +1,35 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "../ckpts/Meta-Llama-3-8B-Instruct",
+  "bias": "none",
+  "fan_in_fan_out": false,
+  "inference_mode": true,
+  "init_lora_weights": true,
+  "layer_replication": null,
+  "layers_pattern": null,
+  "layers_to_transform": null,
+  "loftq_config": {},
+  "lora_alpha": 32,
+  "lora_dropout": 0.1,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 16,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "gate_proj",
+    "q_proj",
+    "lm_head",
+    "o_proj",
+    "v_proj",
+    "up_proj",
+    "k_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-300/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a21bde55efc0304e99e5e82e98c55a75914860b89ddab4ffc6669646aa38aa13
+size 1138856856

checkpoint-300/global_step300/bf16_zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3915aeacc40b0f3920cafe63f8a83b537514f4711d5e9f12886aa2543e2ce4f
+size 528781328

checkpoint-300/global_step300/mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92259a465b6d29edb36bf9a1e7975102d5bc390e4d97bf2c72e274d222b5d254
+size 199905337

checkpoint-300/latest

@@ -0,0 +1 @@
+global_step300

checkpoint-300/rng_state.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:75bbf68b4afb9b24c52daea43d34c43c94c81e77b7a3e07025e50c711c44a449
+size 14244

checkpoint-300/trainer_state.json

@@ -0,0 +1,2133 @@
+{
+  "best_metric": null,
+  "best_model_checkpoint": null,
+  "epoch": 2.56,
+  "eval_steps": 500,
+  "global_step": 300,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.008533333333333334,
+      "grad_norm": 100.21848203113535,
+      "learning_rate": 0.0,
+      "loss": 7.1962,
+      "step": 1
+    },
+    {
+      "epoch": 0.017066666666666667,
+      "grad_norm": 96.43006188910957,
+      "learning_rate": 3.0102999566398115e-06,
+      "loss": 6.9414,
+      "step": 2
+    },
+    {
+      "epoch": 0.0256,
+      "grad_norm": 97.35803466618715,
+      "learning_rate": 4.771212547196624e-06,
+      "loss": 7.0102,
+      "step": 3
+    },
+    {
+      "epoch": 0.034133333333333335,
+      "grad_norm": 95.14837816372646,
+      "learning_rate": 6.020599913279623e-06,
+      "loss": 6.5295,
+      "step": 4
+    },
+    {
+      "epoch": 0.042666666666666665,
+      "grad_norm": 91.76544275692784,
+      "learning_rate": 6.989700043360187e-06,
+      "loss": 6.4806,
+      "step": 5
+    },
+    {
+      "epoch": 0.0512,
+      "grad_norm": 84.4494318688335,
+      "learning_rate": 7.781512503836437e-06,
+      "loss": 6.4194,
+      "step": 6
+    },
+    {
+      "epoch": 0.05973333333333333,
+      "grad_norm": 71.37977490595638,
+      "learning_rate": 8.450980400142568e-06,
+      "loss": 5.4953,
+      "step": 7
+    },
+    {
+      "epoch": 0.06826666666666667,
+      "grad_norm": 49.31153456754566,
+      "learning_rate": 9.030899869919434e-06,
+      "loss": 5.4123,
+      "step": 8
+    },
+    {
+      "epoch": 0.0768,
+      "grad_norm": 20.37296364560341,
+      "learning_rate": 9.542425094393249e-06,
+      "loss": 5.2334,
+      "step": 9
+    },
+    {
+      "epoch": 0.08533333333333333,
+      "grad_norm": 7.968467079076881,
+      "learning_rate": 9.999999999999999e-06,
+      "loss": 5.0282,
+      "step": 10
+    },
+    {
+      "epoch": 0.09386666666666667,
+      "grad_norm": 3.559446532055649,
+      "learning_rate": 1.041392685158225e-05,
+      "loss": 4.612,
+      "step": 11
+    },
+    {
+      "epoch": 0.1024,
+      "grad_norm": 3.5528846947995674,
+      "learning_rate": 1.0791812460476248e-05,
+      "loss": 4.9475,
+      "step": 12
+    },
+    {
+      "epoch": 0.11093333333333333,
+      "grad_norm": 3.541968897471334,
+      "learning_rate": 1.1139433523068365e-05,
+      "loss": 4.2777,
+      "step": 13
+    },
+    {
+      "epoch": 0.11946666666666667,
+      "grad_norm": 3.54718070036198,
+      "learning_rate": 1.1461280356782378e-05,
+      "loss": 4.3507,
+      "step": 14
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 3.8632334830606747,
+      "learning_rate": 1.1760912590556813e-05,
+      "loss": 4.5364,
+      "step": 15
+    },
+    {
+      "epoch": 0.13653333333333334,
+      "grad_norm": 3.6637424744054004,
+      "learning_rate": 1.2041199826559246e-05,
+      "loss": 3.9672,
+      "step": 16
+    },
+    {
+      "epoch": 0.14506666666666668,
+      "grad_norm": 3.919802904818311,
+      "learning_rate": 1.230448921378274e-05,
+      "loss": 4.0618,
+      "step": 17
+    },
+    {
+      "epoch": 0.1536,
+      "grad_norm": 4.71904950738746,
+      "learning_rate": 1.2552725051033058e-05,
+      "loss": 4.6656,
+      "step": 18
+    },
+    {
+      "epoch": 0.16213333333333332,
+      "grad_norm": 4.6656317698690835,
+      "learning_rate": 1.2787536009528288e-05,
+      "loss": 4.1131,
+      "step": 19
+    },
+    {
+      "epoch": 0.17066666666666666,
+      "grad_norm": 5.145138692367417,
+      "learning_rate": 1.301029995663981e-05,
+      "loss": 4.0989,
+      "step": 20
+    },
+    {
+      "epoch": 0.1792,
+      "grad_norm": 5.923538014759818,
+      "learning_rate": 1.3222192947339192e-05,
+      "loss": 4.4991,
+      "step": 21
+    },
+    {
+      "epoch": 0.18773333333333334,
+      "grad_norm": 5.941056962941364,
+      "learning_rate": 1.3424226808222062e-05,
+      "loss": 4.0836,
+      "step": 22
+    },
+    {
+      "epoch": 0.19626666666666667,
+      "grad_norm": 6.171026012117947,
+      "learning_rate": 1.3617278360175927e-05,
+      "loss": 3.6861,
+      "step": 23
+    },
+    {
+      "epoch": 0.2048,
+      "grad_norm": 7.130542138930838,
+      "learning_rate": 1.380211241711606e-05,
+      "loss": 4.0958,
+      "step": 24
+    },
+    {
+      "epoch": 0.21333333333333335,
+      "grad_norm": 7.328837606110418,
+      "learning_rate": 1.3979400086720374e-05,
+      "loss": 3.9524,
+      "step": 25
+    },
+    {
+      "epoch": 0.22186666666666666,
+      "grad_norm": 6.923489005711429,
+      "learning_rate": 1.4149733479708178e-05,
+      "loss": 3.6062,
+      "step": 26
+    },
+    {
+      "epoch": 0.2304,
+      "grad_norm": 6.409498926059221,
+      "learning_rate": 1.4313637641589872e-05,
+      "loss": 3.2034,
+      "step": 27
+    },
+    {
+      "epoch": 0.23893333333333333,
+      "grad_norm": 5.781628405584682,
+      "learning_rate": 1.4471580313422191e-05,
+      "loss": 2.8158,
+      "step": 28
+    },
+    {
+      "epoch": 0.24746666666666667,
+      "grad_norm": 6.2927590068243315,
+      "learning_rate": 1.4623979978989559e-05,
+      "loss": 2.9803,
+      "step": 29
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 6.103844678630006,
+      "learning_rate": 1.4771212547196623e-05,
+      "loss": 2.847,
+      "step": 30
+    },
+    {
+      "epoch": 0.26453333333333334,
+      "grad_norm": 7.656341925867144,
+      "learning_rate": 1.4913616938342726e-05,
+      "loss": 3.0907,
+      "step": 31
+    },
+    {
+      "epoch": 0.2730666666666667,
+      "grad_norm": 6.324242877501844,
+      "learning_rate": 1.5051499783199059e-05,
+      "loss": 2.3467,
+      "step": 32
+    },
+    {
+      "epoch": 0.2816,
+      "grad_norm": 7.606313101162739,
+      "learning_rate": 1.5185139398778874e-05,
+      "loss": 2.5292,
+      "step": 33
+    },
+    {
+      "epoch": 0.29013333333333335,
+      "grad_norm": 8.553792493849265,
+      "learning_rate": 1.531478917042255e-05,
+      "loss": 2.4547,
+      "step": 34
+    },
+    {
+      "epoch": 0.2986666666666667,
+      "grad_norm": 8.483368703272543,
+      "learning_rate": 1.5440680443502753e-05,
+      "loss": 2.1956,
+      "step": 35
+    },
+    {
+      "epoch": 0.3072,
+      "grad_norm": 8.364739352838077,
+      "learning_rate": 1.5563025007672873e-05,
+      "loss": 1.8552,
+      "step": 36
+    },
+    {
+      "epoch": 0.3157333333333333,
+      "grad_norm": 9.37663682000104,
+      "learning_rate": 1.5682017240669948e-05,
+      "loss": 1.9228,
+      "step": 37
+    },
+    {
+      "epoch": 0.32426666666666665,
+      "grad_norm": 9.823047193440066,
+      "learning_rate": 1.57978359661681e-05,
+      "loss": 1.7033,
+      "step": 38
+    },
+    {
+      "epoch": 0.3328,
+      "grad_norm": 9.692618512955894,
+      "learning_rate": 1.591064607026499e-05,
+      "loss": 1.3768,
+      "step": 39
+    },
+    {
+      "epoch": 0.3413333333333333,
+      "grad_norm": 9.08889021911031,
+      "learning_rate": 1.6020599913279622e-05,
+      "loss": 1.3015,
+      "step": 40
+    },
+    {
+      "epoch": 0.34986666666666666,
+      "grad_norm": 8.081534221516058,
+      "learning_rate": 1.6127838567197353e-05,
+      "loss": 0.9228,
+      "step": 41
+    },
+    {
+      "epoch": 0.3584,
+      "grad_norm": 6.238638048950311,
+      "learning_rate": 1.6232492903979005e-05,
+      "loss": 0.7267,
+      "step": 42
+    },
+    {
+      "epoch": 0.36693333333333333,
+      "grad_norm": 3.4058036861773604,
+      "learning_rate": 1.6334684555795865e-05,
+      "loss": 0.5875,
+      "step": 43
+    },
+    {
+      "epoch": 0.37546666666666667,
+      "grad_norm": 2.079163829467713,
+      "learning_rate": 1.6434526764861872e-05,
+      "loss": 0.6355,
+      "step": 44
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 1.5597487833024746,
+      "learning_rate": 1.6532125137753435e-05,
+      "loss": 0.5106,
+      "step": 45
+    },
+    {
+      "epoch": 0.39253333333333335,
+      "grad_norm": 2.491689602375256,
+      "learning_rate": 1.662757831681574e-05,
+      "loss": 0.6454,
+      "step": 46
+    },
+    {
+      "epoch": 0.4010666666666667,
+      "grad_norm": 2.010880438195854,
+      "learning_rate": 1.672097857935717e-05,
+      "loss": 0.4757,
+      "step": 47
+    },
+    {
+      "epoch": 0.4096,
+      "grad_norm": 1.9452805114322096,
+      "learning_rate": 1.681241237375587e-05,
+      "loss": 0.4133,
+      "step": 48
+    },
+    {
+      "epoch": 0.41813333333333336,
+      "grad_norm": 1.7620848552306103,
+      "learning_rate": 1.6901960800285137e-05,
+      "loss": 0.4004,
+      "step": 49
+    },
+    {
+      "epoch": 0.4266666666666667,
+      "grad_norm": 1.278224489774809,
+      "learning_rate": 1.6989700043360187e-05,
+      "loss": 0.3523,
+      "step": 50
+    },
+    {
+      "epoch": 0.4352,
+      "grad_norm": 1.6151354758303231,
+      "learning_rate": 1.7075701760979363e-05,
+      "loss": 0.4317,
+      "step": 51
+    },
+    {
+      "epoch": 0.4437333333333333,
+      "grad_norm": 1.3451396055695035,
+      "learning_rate": 1.716003343634799e-05,
+      "loss": 0.3474,
+      "step": 52
+    },
+    {
+      "epoch": 0.45226666666666665,
+      "grad_norm": 1.6814977782362666,
+      "learning_rate": 1.724275869600789e-05,
+      "loss": 0.3706,
+      "step": 53
+    },
+    {
+      "epoch": 0.4608,
+      "grad_norm": 1.1682442432688667,
+      "learning_rate": 1.7323937598229687e-05,
+      "loss": 0.3488,
+      "step": 54
+    },
+    {
+      "epoch": 0.4693333333333333,
+      "grad_norm": 0.8839814540462471,
+      "learning_rate": 1.7403626894942437e-05,
+      "loss": 0.293,
+      "step": 55
+    },
+    {
+      "epoch": 0.47786666666666666,
+      "grad_norm": 0.7974304806295485,
+      "learning_rate": 1.7481880270062003e-05,
+      "loss": 0.2717,
+      "step": 56
+    },
+    {
+      "epoch": 0.4864,
+      "grad_norm": 0.6232203657585239,
+      "learning_rate": 1.7558748556724913e-05,
+      "loss": 0.1741,
+      "step": 57
+    },
+    {
+      "epoch": 0.49493333333333334,
+      "grad_norm": 0.6850688604563008,
+      "learning_rate": 1.763427993562937e-05,
+      "loss": 0.228,
+      "step": 58
+    },
+    {
+      "epoch": 0.5034666666666666,
+      "grad_norm": 0.5923826384300431,
+      "learning_rate": 1.7708520116421443e-05,
+      "loss": 0.2131,
+      "step": 59
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.489754430485032,
+      "learning_rate": 1.7781512503836432e-05,
+      "loss": 0.165,
+      "step": 60
+    },
+    {
+      "epoch": 0.5205333333333333,
+      "grad_norm": 0.5280940052395061,
+      "learning_rate": 1.7853298350107667e-05,
+      "loss": 0.1658,
+      "step": 61
+    },
+    {
+      "epoch": 0.5290666666666667,
+      "grad_norm": 0.4750905992036739,
+      "learning_rate": 1.7923916894982537e-05,
+      "loss": 0.1438,
+      "step": 62
+    },
+    {
+      "epoch": 0.5376,
+      "grad_norm": 0.4162998392722401,
+      "learning_rate": 1.7993405494535815e-05,
+      "loss": 0.1555,
+      "step": 63
+    },
+    {
+      "epoch": 0.5461333333333334,
+      "grad_norm": 0.26863266246370443,
+      "learning_rate": 1.806179973983887e-05,
+      "loss": 0.1323,
+      "step": 64
+    },
+    {
+      "epoch": 0.5546666666666666,
+      "grad_norm": 0.26534221125601215,
+      "learning_rate": 1.8129133566428553e-05,
+      "loss": 0.1671,
+      "step": 65
+    },
+    {
+      "epoch": 0.5632,
+      "grad_norm": 0.2548662962257576,
+      "learning_rate": 1.8195439355418686e-05,
+      "loss": 0.1308,
+      "step": 66
+    },
+    {
+      "epoch": 0.5717333333333333,
+      "grad_norm": 0.18045394638578796,
+      "learning_rate": 1.8260748027008263e-05,
+      "loss": 0.1262,
+      "step": 67
+    },
+    {
+      "epoch": 0.5802666666666667,
+      "grad_norm": 0.17070388073612064,
+      "learning_rate": 1.8325089127062364e-05,
+      "loss": 0.1192,
+      "step": 68
+    },
+    {
+      "epoch": 0.5888,
+      "grad_norm": 0.1531381679244776,
+      "learning_rate": 1.8388490907372553e-05,
+      "loss": 0.1274,
+      "step": 69
+    },
+    {
+      "epoch": 0.5973333333333334,
+      "grad_norm": 0.18196434993687946,
+      "learning_rate": 1.8450980400142568e-05,
+      "loss": 0.1375,
+      "step": 70
+    },
+    {
+      "epoch": 0.6058666666666667,
+      "grad_norm": 0.15324416972951205,
+      "learning_rate": 1.8512583487190752e-05,
+      "loss": 0.1599,
+      "step": 71
+    },
+    {
+      "epoch": 0.6144,
+      "grad_norm": 0.10884462064503801,
+      "learning_rate": 1.857332496431268e-05,
+      "loss": 0.1041,
+      "step": 72
+    },
+    {
+      "epoch": 0.6229333333333333,
+      "grad_norm": 0.12915133528192668,
+      "learning_rate": 1.8633228601204555e-05,
+      "loss": 0.1406,
+      "step": 73
+    },
+    {
+      "epoch": 0.6314666666666666,
+      "grad_norm": 0.12553425699952878,
+      "learning_rate": 1.8692317197309763e-05,
+      "loss": 0.1256,
+      "step": 74
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.11976960918968543,
+      "learning_rate": 1.8750612633916997e-05,
+      "loss": 0.1144,
+      "step": 75
+    },
+    {
+      "epoch": 0.6485333333333333,
+      "grad_norm": 0.115805998298789,
+      "learning_rate": 1.8808135922807914e-05,
+      "loss": 0.1528,
+      "step": 76
+    },
+    {
+      "epoch": 0.6570666666666667,
+      "grad_norm": 0.10325948496697443,
+      "learning_rate": 1.8864907251724818e-05,
+      "loss": 0.1044,
+      "step": 77
+    },
+    {
+      "epoch": 0.6656,
+      "grad_norm": 0.09595064346541006,
+      "learning_rate": 1.8920946026904802e-05,
+      "loss": 0.1534,
+      "step": 78
+    },
+    {
+      "epoch": 0.6741333333333334,
+      "grad_norm": 0.08796742845240496,
+      "learning_rate": 1.8976270912904414e-05,
+      "loss": 0.1155,
+      "step": 79
+    },
+    {
+      "epoch": 0.6826666666666666,
+      "grad_norm": 0.08218991738379527,
+      "learning_rate": 1.9030899869919434e-05,
+      "loss": 0.1311,
+      "step": 80
+    },
+    {
+      "epoch": 0.6912,
+      "grad_norm": 0.08290815261109215,
+      "learning_rate": 1.9084850188786497e-05,
+      "loss": 0.11,
+      "step": 81
+    },
+    {
+      "epoch": 0.6997333333333333,
+      "grad_norm": 0.0794132180176064,
+      "learning_rate": 1.9138138523837165e-05,
+      "loss": 0.1135,
+      "step": 82
+    },
+    {
+      "epoch": 0.7082666666666667,
+      "grad_norm": 0.06934410705255296,
+      "learning_rate": 1.919078092376074e-05,
+      "loss": 0.109,
+      "step": 83
+    },
+    {
+      "epoch": 0.7168,
+      "grad_norm": 0.09000563031870593,
+      "learning_rate": 1.9242792860618813e-05,
+      "loss": 0.12,
+      "step": 84
+    },
+    {
+      "epoch": 0.7253333333333334,
+      "grad_norm": 0.1134042277879818,
+      "learning_rate": 1.929418925714293e-05,
+      "loss": 0.1223,
+      "step": 85
+    },
+    {
+      "epoch": 0.7338666666666667,
+      "grad_norm": 0.09118764690233076,
+      "learning_rate": 1.9344984512435673e-05,
+      "loss": 0.1459,
+      "step": 86
+    },
+    {
+      "epoch": 0.7424,
+      "grad_norm": 0.07873016754353963,
+      "learning_rate": 1.9395192526186183e-05,
+      "loss": 0.1422,
+      "step": 87
+    },
+    {
+      "epoch": 0.7509333333333333,
+      "grad_norm": 0.1796495874463076,
+      "learning_rate": 1.9444826721501687e-05,
+      "loss": 0.1291,
+      "step": 88
+    },
+    {
+      "epoch": 0.7594666666666666,
+      "grad_norm": 0.0679589944174269,
+      "learning_rate": 1.9493900066449125e-05,
+      "loss": 0.108,
+      "step": 89
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.08174688574235538,
+      "learning_rate": 1.9542425094393246e-05,
+      "loss": 0.1081,
+      "step": 90
+    },
+    {
+      "epoch": 0.7765333333333333,
+      "grad_norm": 0.057137370501406756,
+      "learning_rate": 1.9590413923210934e-05,
+      "loss": 0.0934,
+      "step": 91
+    },
+    {
+      "epoch": 0.7850666666666667,
+      "grad_norm": 0.06578111924908255,
+      "learning_rate": 1.9637878273455555e-05,
+      "loss": 0.1085,
+      "step": 92
+    },
+    {
+      "epoch": 0.7936,
+      "grad_norm": 0.08945990540906254,
+      "learning_rate": 1.968482948553935e-05,
+      "loss": 0.1747,
+      "step": 93
+    },
+    {
+      "epoch": 0.8021333333333334,
+      "grad_norm": 0.06183863311044229,
+      "learning_rate": 1.9731278535996986e-05,
+      "loss": 0.1136,
+      "step": 94
+    },
+    {
+      "epoch": 0.8106666666666666,
+      "grad_norm": 0.05777899602544702,
+      "learning_rate": 1.9777236052888476e-05,
+      "loss": 0.0984,
+      "step": 95
+    },
+    {
+      "epoch": 0.8192,
+      "grad_norm": 0.08130851607693534,
+      "learning_rate": 1.9822712330395683e-05,
+      "loss": 0.187,
+      "step": 96
+    },
+    {
+      "epoch": 0.8277333333333333,
+      "grad_norm": 0.06426546202002927,
+      "learning_rate": 1.986771734266245e-05,
+      "loss": 0.1296,
+      "step": 97
+    },
+    {
+      "epoch": 0.8362666666666667,
+      "grad_norm": 0.069692313707994,
+      "learning_rate": 1.991226075692495e-05,
+      "loss": 0.1404,
+      "step": 98
+    },
+    {
+      "epoch": 0.8448,
+      "grad_norm": 0.05494542266886729,
+      "learning_rate": 1.9956351945975496e-05,
+      "loss": 0.116,
+      "step": 99
+    },
+    {
+      "epoch": 0.8533333333333334,
+      "grad_norm": 0.07571686966840627,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 0.1539,
+      "step": 100
+    },
+    {
+      "epoch": 0.8618666666666667,
+      "grad_norm": 0.054351059117603705,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 101
+    },
+    {
+      "epoch": 0.8704,
+      "grad_norm": 0.06531899611551092,
+      "learning_rate": 2e-05,
+      "loss": 0.0827,
+      "step": 102
+    },
+    {
+      "epoch": 0.8789333333333333,
+      "grad_norm": 0.06131678646504652,
+      "learning_rate": 2e-05,
+      "loss": 0.1266,
+      "step": 103
+    },
+    {
+      "epoch": 0.8874666666666666,
+      "grad_norm": 0.06850220540661824,
+      "learning_rate": 2e-05,
+      "loss": 0.1456,
+      "step": 104
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.05806908951252483,
+      "learning_rate": 2e-05,
+      "loss": 0.0954,
+      "step": 105
+    },
+    {
+      "epoch": 0.9045333333333333,
+      "grad_norm": 0.06503642452033717,
+      "learning_rate": 2e-05,
+      "loss": 0.1417,
+      "step": 106
+    },
+    {
+      "epoch": 0.9130666666666667,
+      "grad_norm": 0.050486271853277066,
+      "learning_rate": 2e-05,
+      "loss": 0.0959,
+      "step": 107
+    },
+    {
+      "epoch": 0.9216,
+      "grad_norm": 0.07746063813802379,
+      "learning_rate": 2e-05,
+      "loss": 0.1256,
+      "step": 108
+    },
+    {
+      "epoch": 0.9301333333333334,
+      "grad_norm": 0.051231172380840004,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 109
+    },
+    {
+      "epoch": 0.9386666666666666,
+      "grad_norm": 0.056296443557859455,
+      "learning_rate": 2e-05,
+      "loss": 0.1056,
+      "step": 110
+    },
+    {
+      "epoch": 0.9472,
+      "grad_norm": 0.05058663240713958,
+      "learning_rate": 2e-05,
+      "loss": 0.0971,
+      "step": 111
+    },
+    {
+      "epoch": 0.9557333333333333,
+      "grad_norm": 0.05532886570130611,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 112
+    },
+    {
+      "epoch": 0.9642666666666667,
+      "grad_norm": 0.05327811326654907,
+      "learning_rate": 2e-05,
+      "loss": 0.0989,
+      "step": 113
+    },
+    {
+      "epoch": 0.9728,
+      "grad_norm": 0.05663279364147864,
+      "learning_rate": 2e-05,
+      "loss": 0.0958,
+      "step": 114
+    },
+    {
+      "epoch": 0.9813333333333333,
+      "grad_norm": 0.04930904541225805,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 115
+    },
+    {
+      "epoch": 0.9898666666666667,
+      "grad_norm": 0.06096947951115022,
+      "learning_rate": 2e-05,
+      "loss": 0.106,
+      "step": 116
+    },
+    {
+      "epoch": 0.9984,
+      "grad_norm": 0.050092322361182495,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 117
+    },
+    {
+      "epoch": 1.0069333333333332,
+      "grad_norm": 0.04980408443758999,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 118
+    },
+    {
+      "epoch": 1.0154666666666667,
+      "grad_norm": 0.051183082721834305,
+      "learning_rate": 2e-05,
+      "loss": 0.1049,
+      "step": 119
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.04332220265802814,
+      "learning_rate": 2e-05,
+      "loss": 0.0983,
+      "step": 120
+    },
+    {
+      "epoch": 1.0325333333333333,
+      "grad_norm": 0.07211731499677299,
+      "learning_rate": 2e-05,
+      "loss": 0.1386,
+      "step": 121
+    },
+    {
+      "epoch": 1.0410666666666666,
+      "grad_norm": 0.06550870223740553,
+      "learning_rate": 2e-05,
+      "loss": 0.1334,
+      "step": 122
+    },
+    {
+      "epoch": 1.0496,
+      "grad_norm": 0.05331839690767287,
+      "learning_rate": 2e-05,
+      "loss": 0.1014,
+      "step": 123
+    },
+    {
+      "epoch": 1.0581333333333334,
+      "grad_norm": 0.05227685628767905,
+      "learning_rate": 2e-05,
+      "loss": 0.1098,
+      "step": 124
+    },
+    {
+      "epoch": 1.0666666666666667,
+      "grad_norm": 0.07641693882491171,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 125
+    },
+    {
+      "epoch": 1.0752,
+      "grad_norm": 0.052835367770791786,
+      "learning_rate": 2e-05,
+      "loss": 0.1231,
+      "step": 126
+    },
+    {
+      "epoch": 1.0837333333333334,
+      "grad_norm": 0.07520329755025788,
+      "learning_rate": 2e-05,
+      "loss": 0.085,
+      "step": 127
+    },
+    {
+      "epoch": 1.0922666666666667,
+      "grad_norm": 0.07670066152157425,
+      "learning_rate": 2e-05,
+      "loss": 0.1071,
+      "step": 128
+    },
+    {
+      "epoch": 1.1008,
+      "grad_norm": 0.052832906560645154,
+      "learning_rate": 2e-05,
+      "loss": 0.1093,
+      "step": 129
+    },
+    {
+      "epoch": 1.1093333333333333,
+      "grad_norm": 0.06573889037311398,
+      "learning_rate": 2e-05,
+      "loss": 0.1193,
+      "step": 130
+    },
+    {
+      "epoch": 1.1178666666666666,
+      "grad_norm": 0.05175471296566334,
+      "learning_rate": 2e-05,
+      "loss": 0.1184,
+      "step": 131
+    },
+    {
+      "epoch": 1.1264,
+      "grad_norm": 0.05912231419793496,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 132
+    },
+    {
+      "epoch": 1.1349333333333333,
+      "grad_norm": 0.04899140475981105,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 133
+    },
+    {
+      "epoch": 1.1434666666666666,
+      "grad_norm": 0.05939916939142137,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 134
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.0516819217599706,
+      "learning_rate": 2e-05,
+      "loss": 0.0834,
+      "step": 135
+    },
+    {
+      "epoch": 1.1605333333333334,
+      "grad_norm": 0.05456440346454737,
+      "learning_rate": 2e-05,
+      "loss": 0.1183,
+      "step": 136
+    },
+    {
+      "epoch": 1.1690666666666667,
+      "grad_norm": 0.059906464476343235,
+      "learning_rate": 2e-05,
+      "loss": 0.1048,
+      "step": 137
+    },
+    {
+      "epoch": 1.1776,
+      "grad_norm": 0.0720112680204319,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 138
+    },
+    {
+      "epoch": 1.1861333333333333,
+      "grad_norm": 0.04940202805828527,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 139
+    },
+    {
+      "epoch": 1.1946666666666665,
+      "grad_norm": 0.060088609545130046,
+      "learning_rate": 2e-05,
+      "loss": 0.0952,
+      "step": 140
+    },
+    {
+      "epoch": 1.2032,
+      "grad_norm": 0.04694761423612446,
+      "learning_rate": 2e-05,
+      "loss": 0.0717,
+      "step": 141
+    },
+    {
+      "epoch": 1.2117333333333333,
+      "grad_norm": 0.05628581562512457,
+      "learning_rate": 2e-05,
+      "loss": 0.1062,
+      "step": 142
+    },
+    {
+      "epoch": 1.2202666666666666,
+      "grad_norm": 0.06876420990437652,
+      "learning_rate": 2e-05,
+      "loss": 0.1218,
+      "step": 143
+    },
+    {
+      "epoch": 1.2288000000000001,
+      "grad_norm": 0.058774700501610655,
+      "learning_rate": 2e-05,
+      "loss": 0.1125,
+      "step": 144
+    },
+    {
+      "epoch": 1.2373333333333334,
+      "grad_norm": 0.061855922064341186,
+      "learning_rate": 2e-05,
+      "loss": 0.1295,
+      "step": 145
+    },
+    {
+      "epoch": 1.2458666666666667,
+      "grad_norm": 0.0813047704730138,
+      "learning_rate": 2e-05,
+      "loss": 0.1165,
+      "step": 146
+    },
+    {
+      "epoch": 1.2544,
+      "grad_norm": 0.061374000305305446,
+      "learning_rate": 2e-05,
+      "loss": 0.1094,
+      "step": 147
+    },
+    {
+      "epoch": 1.2629333333333332,
+      "grad_norm": 0.055537169110833,
+      "learning_rate": 2e-05,
+      "loss": 0.1054,
+      "step": 148
+    },
+    {
+      "epoch": 1.2714666666666667,
+      "grad_norm": 0.04423248714119304,
+      "learning_rate": 2e-05,
+      "loss": 0.0841,
+      "step": 149
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.049931966607835034,
+      "learning_rate": 2e-05,
+      "loss": 0.0961,
+      "step": 150
+    },
+    {
+      "epoch": 1.2885333333333333,
+      "grad_norm": 0.06178656953298769,
+      "learning_rate": 2e-05,
+      "loss": 0.0854,
+      "step": 151
+    },
+    {
+      "epoch": 1.2970666666666666,
+      "grad_norm": 0.05783812343287897,
+      "learning_rate": 2e-05,
+      "loss": 0.1141,
+      "step": 152
+    },
+    {
+      "epoch": 1.3056,
+      "grad_norm": 0.048955120400167584,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 153
+    },
+    {
+      "epoch": 1.3141333333333334,
+      "grad_norm": 0.12270174746806978,
+      "learning_rate": 2e-05,
+      "loss": 0.1553,
+      "step": 154
+    },
+    {
+      "epoch": 1.3226666666666667,
+      "grad_norm": 0.06928026959973474,
+      "learning_rate": 2e-05,
+      "loss": 0.1274,
+      "step": 155
+    },
+    {
+      "epoch": 1.3312,
+      "grad_norm": 0.04756100666105405,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 156
+    },
+    {
+      "epoch": 1.3397333333333332,
+      "grad_norm": 0.056054951338196934,
+      "learning_rate": 2e-05,
+      "loss": 0.0831,
+      "step": 157
+    },
+    {
+      "epoch": 1.3482666666666667,
+      "grad_norm": 0.0516990471964239,
+      "learning_rate": 2e-05,
+      "loss": 0.0883,
+      "step": 158
+    },
+    {
+      "epoch": 1.3568,
+      "grad_norm": 0.06011650542069954,
+      "learning_rate": 2e-05,
+      "loss": 0.0938,
+      "step": 159
+    },
+    {
+      "epoch": 1.3653333333333333,
+      "grad_norm": 0.051831307951873976,
+      "learning_rate": 2e-05,
+      "loss": 0.1019,
+      "step": 160
+    },
+    {
+      "epoch": 1.3738666666666668,
+      "grad_norm": 0.0544902702048936,
+      "learning_rate": 2e-05,
+      "loss": 0.0906,
+      "step": 161
+    },
+    {
+      "epoch": 1.3824,
+      "grad_norm": 0.06057617032526,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 162
+    },
+    {
+      "epoch": 1.3909333333333334,
+      "grad_norm": 0.07288058025189605,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 163
+    },
+    {
+      "epoch": 1.3994666666666666,
+      "grad_norm": 0.05639043792084219,
+      "learning_rate": 2e-05,
+      "loss": 0.1031,
+      "step": 164
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.0586469408837505,
+      "learning_rate": 2e-05,
+      "loss": 0.1143,
+      "step": 165
+    },
+    {
+      "epoch": 1.4165333333333332,
+      "grad_norm": 0.059243429040783904,
+      "learning_rate": 2e-05,
+      "loss": 0.0838,
+      "step": 166
+    },
+    {
+      "epoch": 1.4250666666666667,
+      "grad_norm": 0.0621476448363388,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 167
+    },
+    {
+      "epoch": 1.4336,
+      "grad_norm": 0.06083867682720169,
+      "learning_rate": 2e-05,
+      "loss": 0.1119,
+      "step": 168
+    },
+    {
+      "epoch": 1.4421333333333333,
+      "grad_norm": 0.09583165335305677,
+      "learning_rate": 2e-05,
+      "loss": 0.1028,
+      "step": 169
+    },
+    {
+      "epoch": 1.4506666666666668,
+      "grad_norm": 0.06411638581314043,
+      "learning_rate": 2e-05,
+      "loss": 0.1181,
+      "step": 170
+    },
+    {
+      "epoch": 1.4592,
+      "grad_norm": 0.05632977234908945,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 171
+    },
+    {
+      "epoch": 1.4677333333333333,
+      "grad_norm": 0.05666068971337413,
+      "learning_rate": 2e-05,
+      "loss": 0.1116,
+      "step": 172
+    },
+    {
+      "epoch": 1.4762666666666666,
+      "grad_norm": 0.04502062723807536,
+      "learning_rate": 2e-05,
+      "loss": 0.0588,
+      "step": 173
+    },
+    {
+      "epoch": 1.4848,
+      "grad_norm": 0.05916500176868301,
+      "learning_rate": 2e-05,
+      "loss": 0.0949,
+      "step": 174
+    },
+    {
+      "epoch": 1.4933333333333334,
+      "grad_norm": 0.056484273808864845,
+      "learning_rate": 2e-05,
+      "loss": 0.0948,
+      "step": 175
+    },
+    {
+      "epoch": 1.5018666666666667,
+      "grad_norm": 0.06652084448571674,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 176
+    },
+    {
+      "epoch": 1.5104,
+      "grad_norm": 0.05400238078068516,
+      "learning_rate": 2e-05,
+      "loss": 0.0919,
+      "step": 177
+    },
+    {
+      "epoch": 1.5189333333333335,
+      "grad_norm": 0.04979579743346662,
+      "learning_rate": 2e-05,
+      "loss": 0.0879,
+      "step": 178
+    },
+    {
+      "epoch": 1.5274666666666668,
+      "grad_norm": 0.06876105414733971,
+      "learning_rate": 2e-05,
+      "loss": 0.1162,
+      "step": 179
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.05633262015499721,
+      "learning_rate": 2e-05,
+      "loss": 0.1142,
+      "step": 180
+    },
+    {
+      "epoch": 1.5445333333333333,
+      "grad_norm": 0.0599508967519892,
+      "learning_rate": 2e-05,
+      "loss": 0.1073,
+      "step": 181
+    },
+    {
+      "epoch": 1.5530666666666666,
+      "grad_norm": 0.058912170976454126,
+      "learning_rate": 2e-05,
+      "loss": 0.1102,
+      "step": 182
+    },
+    {
+      "epoch": 1.5615999999999999,
+      "grad_norm": 0.05361414988566254,
+      "learning_rate": 2e-05,
+      "loss": 0.0885,
+      "step": 183
+    },
+    {
+      "epoch": 1.5701333333333334,
+      "grad_norm": 0.04517277847384077,
+      "learning_rate": 2e-05,
+      "loss": 0.0763,
+      "step": 184
+    },
+    {
+      "epoch": 1.5786666666666667,
+      "grad_norm": 0.05551553662051771,
+      "learning_rate": 2e-05,
+      "loss": 0.0877,
+      "step": 185
+    },
+    {
+      "epoch": 1.5872000000000002,
+      "grad_norm": 0.05814223969236194,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 186
+    },
+    {
+      "epoch": 1.5957333333333334,
+      "grad_norm": 0.05710054166191597,
+      "learning_rate": 2e-05,
+      "loss": 0.0962,
+      "step": 187
+    },
+    {
+      "epoch": 1.6042666666666667,
+      "grad_norm": 0.054744343932104075,
+      "learning_rate": 2e-05,
+      "loss": 0.0873,
+      "step": 188
+    },
+    {
+      "epoch": 1.6128,
+      "grad_norm": 0.051145521687090995,
+      "learning_rate": 2e-05,
+      "loss": 0.0855,
+      "step": 189
+    },
+    {
+      "epoch": 1.6213333333333333,
+      "grad_norm": 0.05414658860737789,
+      "learning_rate": 2e-05,
+      "loss": 0.0872,
+      "step": 190
+    },
+    {
+      "epoch": 1.6298666666666666,
+      "grad_norm": 0.05084744595533029,
+      "learning_rate": 2e-05,
+      "loss": 0.0891,
+      "step": 191
+    },
+    {
+      "epoch": 1.6383999999999999,
+      "grad_norm": 0.0567070082820597,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 192
+    },
+    {
+      "epoch": 1.6469333333333334,
+      "grad_norm": 0.0494785311411315,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 193
+    },
+    {
+      "epoch": 1.6554666666666666,
+      "grad_norm": 0.062341158530385396,
+      "learning_rate": 2e-05,
+      "loss": 0.1154,
+      "step": 194
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.059888336716275685,
+      "learning_rate": 2e-05,
+      "loss": 0.1037,
+      "step": 195
+    },
+    {
+      "epoch": 1.6725333333333334,
+      "grad_norm": 0.07346562318829057,
+      "learning_rate": 2e-05,
+      "loss": 0.1329,
+      "step": 196
+    },
+    {
+      "epoch": 1.6810666666666667,
+      "grad_norm": 0.0792360016934733,
+      "learning_rate": 2e-05,
+      "loss": 0.1392,
+      "step": 197
+    },
+    {
+      "epoch": 1.6896,
+      "grad_norm": 0.0800342963229883,
+      "learning_rate": 2e-05,
+      "loss": 0.1199,
+      "step": 198
+    },
+    {
+      "epoch": 1.6981333333333333,
+      "grad_norm": 0.06848045933195548,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 199
+    },
+    {
+      "epoch": 1.7066666666666666,
+      "grad_norm": 0.05743199019316764,
+      "learning_rate": 2e-05,
+      "loss": 0.0811,
+      "step": 200
+    },
+    {
+      "epoch": 1.7151999999999998,
+      "grad_norm": 0.07170531168284446,
+      "learning_rate": 2e-05,
+      "loss": 0.1079,
+      "step": 201
+    },
+    {
+      "epoch": 1.7237333333333333,
+      "grad_norm": 0.05772905481368506,
+      "learning_rate": 2e-05,
+      "loss": 0.0844,
+      "step": 202
+    },
+    {
+      "epoch": 1.7322666666666666,
+      "grad_norm": 0.07504946014098464,
+      "learning_rate": 2e-05,
+      "loss": 0.1257,
+      "step": 203
+    },
+    {
+      "epoch": 1.7408000000000001,
+      "grad_norm": 0.06450179819785838,
+      "learning_rate": 2e-05,
+      "loss": 0.1104,
+      "step": 204
+    },
+    {
+      "epoch": 1.7493333333333334,
+      "grad_norm": 0.06595445965110332,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 205
+    },
+    {
+      "epoch": 1.7578666666666667,
+      "grad_norm": 0.07203558121131749,
+      "learning_rate": 2e-05,
+      "loss": 0.1117,
+      "step": 206
+    },
+    {
+      "epoch": 1.7664,
+      "grad_norm": 0.05954646782409283,
+      "learning_rate": 2e-05,
+      "loss": 0.0729,
+      "step": 207
+    },
+    {
+      "epoch": 1.7749333333333333,
+      "grad_norm": 0.06624894584410884,
+      "learning_rate": 2e-05,
+      "loss": 0.0998,
+      "step": 208
+    },
+    {
+      "epoch": 1.7834666666666665,
+      "grad_norm": 0.06888562028256219,
+      "learning_rate": 2e-05,
+      "loss": 0.1398,
+      "step": 209
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.061224088077794406,
+      "learning_rate": 2e-05,
+      "loss": 0.1112,
+      "step": 210
+    },
+    {
+      "epoch": 1.8005333333333333,
+      "grad_norm": 0.06857358883856608,
+      "learning_rate": 2e-05,
+      "loss": 0.1293,
+      "step": 211
+    },
+    {
+      "epoch": 1.8090666666666668,
+      "grad_norm": 0.06177352416779139,
+      "learning_rate": 2e-05,
+      "loss": 0.0884,
+      "step": 212
+    },
+    {
+      "epoch": 1.8176,
+      "grad_norm": 0.08324567429925228,
+      "learning_rate": 2e-05,
+      "loss": 0.1127,
+      "step": 213
+    },
+    {
+      "epoch": 1.8261333333333334,
+      "grad_norm": 0.06771677297787752,
+      "learning_rate": 2e-05,
+      "loss": 0.089,
+      "step": 214
+    },
+    {
+      "epoch": 1.8346666666666667,
+      "grad_norm": 0.07055754809472485,
+      "learning_rate": 2e-05,
+      "loss": 0.1206,
+      "step": 215
+    },
+    {
+      "epoch": 1.8432,
+      "grad_norm": 0.05856797724392531,
+      "learning_rate": 2e-05,
+      "loss": 0.0893,
+      "step": 216
+    },
+    {
+      "epoch": 1.8517333333333332,
+      "grad_norm": 0.07555286129801597,
+      "learning_rate": 2e-05,
+      "loss": 0.0913,
+      "step": 217
+    },
+    {
+      "epoch": 1.8602666666666665,
+      "grad_norm": 0.09242462538643775,
+      "learning_rate": 2e-05,
+      "loss": 0.1241,
+      "step": 218
+    },
+    {
+      "epoch": 1.8688,
+      "grad_norm": 0.06550805537088374,
+      "learning_rate": 2e-05,
+      "loss": 0.0819,
+      "step": 219
+    },
+    {
+      "epoch": 1.8773333333333333,
+      "grad_norm": 0.06016048263236861,
+      "learning_rate": 2e-05,
+      "loss": 0.0955,
+      "step": 220
+    },
+    {
+      "epoch": 1.8858666666666668,
+      "grad_norm": 0.06856661106001445,
+      "learning_rate": 2e-05,
+      "loss": 0.1132,
+      "step": 221
+    },
+    {
+      "epoch": 1.8944,
+      "grad_norm": 0.06383306455000529,
+      "learning_rate": 2e-05,
+      "loss": 0.1086,
+      "step": 222
+    },
+    {
+      "epoch": 1.9029333333333334,
+      "grad_norm": 0.07240472757239624,
+      "learning_rate": 2e-05,
+      "loss": 0.0863,
+      "step": 223
+    },
+    {
+      "epoch": 1.9114666666666666,
+      "grad_norm": 0.07845654471077741,
+      "learning_rate": 2e-05,
+      "loss": 0.1284,
+      "step": 224
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.07192185833649212,
+      "learning_rate": 2e-05,
+      "loss": 0.1101,
+      "step": 225
+    },
+    {
+      "epoch": 1.9285333333333332,
+      "grad_norm": 0.06390598005596872,
+      "learning_rate": 2e-05,
+      "loss": 0.0917,
+      "step": 226
+    },
+    {
+      "epoch": 1.9370666666666667,
+      "grad_norm": 0.06306138712432224,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 227
+    },
+    {
+      "epoch": 1.9456,
+      "grad_norm": 0.06771381941296478,
+      "learning_rate": 2e-05,
+      "loss": 0.0936,
+      "step": 228
+    },
+    {
+      "epoch": 1.9541333333333335,
+      "grad_norm": 0.05899006803461524,
+      "learning_rate": 2e-05,
+      "loss": 0.0782,
+      "step": 229
+    },
+    {
+      "epoch": 1.9626666666666668,
+      "grad_norm": 0.07426956281950735,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 230
+    },
+    {
+      "epoch": 1.9712,
+      "grad_norm": 0.06567534902293475,
+      "learning_rate": 2e-05,
+      "loss": 0.0956,
+      "step": 231
+    },
+    {
+      "epoch": 1.9797333333333333,
+      "grad_norm": 0.07430395142282198,
+      "learning_rate": 2e-05,
+      "loss": 0.0957,
+      "step": 232
+    },
+    {
+      "epoch": 1.9882666666666666,
+      "grad_norm": 0.05834447367264806,
+      "learning_rate": 2e-05,
+      "loss": 0.0767,
+      "step": 233
+    },
+    {
+      "epoch": 1.9968,
+      "grad_norm": 0.07137090413877054,
+      "learning_rate": 2e-05,
+      "loss": 0.0821,
+      "step": 234
+    },
+    {
+      "epoch": 2.005333333333333,
+      "grad_norm": 0.07797914240805551,
+      "learning_rate": 2e-05,
+      "loss": 0.1175,
+      "step": 235
+    },
+    {
+      "epoch": 2.0138666666666665,
+      "grad_norm": 0.09335648879374187,
+      "learning_rate": 2e-05,
+      "loss": 0.1632,
+      "step": 236
+    },
+    {
+      "epoch": 2.0224,
+      "grad_norm": 0.08280719749100944,
+      "learning_rate": 2e-05,
+      "loss": 0.115,
+      "step": 237
+    },
+    {
+      "epoch": 2.0309333333333335,
+      "grad_norm": 0.08956213539053312,
+      "learning_rate": 2e-05,
+      "loss": 0.109,
+      "step": 238
+    },
+    {
+      "epoch": 2.0394666666666668,
+      "grad_norm": 0.08849786687188893,
+      "learning_rate": 2e-05,
+      "loss": 0.1234,
+      "step": 239
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.16956065138312856,
+      "learning_rate": 2e-05,
+      "loss": 0.1055,
+      "step": 240
+    },
+    {
+      "epoch": 2.0565333333333333,
+      "grad_norm": 0.07377723133015779,
+      "learning_rate": 2e-05,
+      "loss": 0.1068,
+      "step": 241
+    },
+    {
+      "epoch": 2.0650666666666666,
+      "grad_norm": 0.08839651322303756,
+      "learning_rate": 2e-05,
+      "loss": 0.0965,
+      "step": 242
+    },
+    {
+      "epoch": 2.0736,
+      "grad_norm": 0.10641135540308218,
+      "learning_rate": 2e-05,
+      "loss": 0.0887,
+      "step": 243
+    },
+    {
+      "epoch": 2.082133333333333,
+      "grad_norm": 0.09370442613891326,
+      "learning_rate": 2e-05,
+      "loss": 0.0931,
+      "step": 244
+    },
+    {
+      "epoch": 2.0906666666666665,
+      "grad_norm": 0.08988093042206098,
+      "learning_rate": 2e-05,
+      "loss": 0.1201,
+      "step": 245
+    },
+    {
+      "epoch": 2.0992,
+      "grad_norm": 0.07167309696967938,
+      "learning_rate": 2e-05,
+      "loss": 0.0747,
+      "step": 246
+    },
+    {
+      "epoch": 2.1077333333333335,
+      "grad_norm": 0.08822161286881192,
+      "learning_rate": 2e-05,
+      "loss": 0.1044,
+      "step": 247
+    },
+    {
+      "epoch": 2.1162666666666667,
+      "grad_norm": 0.08439713613775747,
+      "learning_rate": 2e-05,
+      "loss": 0.1063,
+      "step": 248
+    },
+    {
+      "epoch": 2.1248,
+      "grad_norm": 0.0942232139678121,
+      "learning_rate": 2e-05,
+      "loss": 0.0865,
+      "step": 249
+    },
+    {
+      "epoch": 2.1333333333333333,
+      "grad_norm": 0.1066306844906037,
+      "learning_rate": 2e-05,
+      "loss": 0.1198,
+      "step": 250
+    },
+    {
+      "epoch": 2.1418666666666666,
+      "grad_norm": 0.10435492421714423,
+      "learning_rate": 2e-05,
+      "loss": 0.0968,
+      "step": 251
+    },
+    {
+      "epoch": 2.1504,
+      "grad_norm": 0.10513719348492025,
+      "learning_rate": 2e-05,
+      "loss": 0.1045,
+      "step": 252
+    },
+    {
+      "epoch": 2.158933333333333,
+      "grad_norm": 0.12580299248756027,
+      "learning_rate": 2e-05,
+      "loss": 0.0941,
+      "step": 253
+    },
+    {
+      "epoch": 2.167466666666667,
+      "grad_norm": 0.10265943183351584,
+      "learning_rate": 2e-05,
+      "loss": 0.0892,
+      "step": 254
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.09473119803029131,
+      "learning_rate": 2e-05,
+      "loss": 0.0721,
+      "step": 255
+    },
+    {
+      "epoch": 2.1845333333333334,
+      "grad_norm": 0.13420988295622846,
+      "learning_rate": 2e-05,
+      "loss": 0.1034,
+      "step": 256
+    },
+    {
+      "epoch": 2.1930666666666667,
+      "grad_norm": 0.11474592330595303,
+      "learning_rate": 2e-05,
+      "loss": 0.1095,
+      "step": 257
+    },
+    {
+      "epoch": 2.2016,
+      "grad_norm": 0.12382914500613644,
+      "learning_rate": 2e-05,
+      "loss": 0.0947,
+      "step": 258
+    },
+    {
+      "epoch": 2.2101333333333333,
+      "grad_norm": 0.12984065755657193,
+      "learning_rate": 2e-05,
+      "loss": 0.0888,
+      "step": 259
+    },
+    {
+      "epoch": 2.2186666666666666,
+      "grad_norm": 0.14210964340218168,
+      "learning_rate": 2e-05,
+      "loss": 0.105,
+      "step": 260
+    },
+    {
+      "epoch": 2.2272,
+      "grad_norm": 0.13862185852250883,
+      "learning_rate": 2e-05,
+      "loss": 0.0876,
+      "step": 261
+    },
+    {
+      "epoch": 2.235733333333333,
+      "grad_norm": 0.17284047904822364,
+      "learning_rate": 2e-05,
+      "loss": 0.0935,
+      "step": 262
+    },
+    {
+      "epoch": 2.244266666666667,
+      "grad_norm": 0.1409465306152598,
+      "learning_rate": 2e-05,
+      "loss": 0.1032,
+      "step": 263
+    },
+    {
+      "epoch": 2.2528,
+      "grad_norm": 0.15551418258816738,
+      "learning_rate": 2e-05,
+      "loss": 0.0871,
+      "step": 264
+    },
+    {
+      "epoch": 2.2613333333333334,
+      "grad_norm": 0.1741617194945442,
+      "learning_rate": 2e-05,
+      "loss": 0.098,
+      "step": 265
+    },
+    {
+      "epoch": 2.2698666666666667,
+      "grad_norm": 0.18387218423978585,
+      "learning_rate": 2e-05,
+      "loss": 0.1168,
+      "step": 266
+    },
+    {
+      "epoch": 2.2784,
+      "grad_norm": 0.21744684782545962,
+      "learning_rate": 2e-05,
+      "loss": 0.1061,
+      "step": 267
+    },
+    {
+      "epoch": 2.2869333333333333,
+      "grad_norm": 0.17372673806896413,
+      "learning_rate": 2e-05,
+      "loss": 0.0927,
+      "step": 268
+    },
+    {
+      "epoch": 2.2954666666666665,
+      "grad_norm": 0.20890400134226653,
+      "learning_rate": 2e-05,
+      "loss": 0.0874,
+      "step": 269
+    },
+    {
+      "epoch": 2.304,
+      "grad_norm": 0.20279172920289834,
+      "learning_rate": 2e-05,
+      "loss": 0.0807,
+      "step": 270
+    },
+    {
+      "epoch": 2.3125333333333336,
+      "grad_norm": 0.23060543242200862,
+      "learning_rate": 2e-05,
+      "loss": 0.0937,
+      "step": 271
+    },
+    {
+      "epoch": 2.321066666666667,
+      "grad_norm": 0.2866337268131343,
+      "learning_rate": 2e-05,
+      "loss": 0.0886,
+      "step": 272
+    },
+    {
+      "epoch": 2.3296,
+      "grad_norm": 0.3203014162518504,
+      "learning_rate": 2e-05,
+      "loss": 0.0741,
+      "step": 273
+    },
+    {
+      "epoch": 2.3381333333333334,
+      "grad_norm": 0.2702455454246614,
+      "learning_rate": 2e-05,
+      "loss": 0.0692,
+      "step": 274
+    },
+    {
+      "epoch": 2.3466666666666667,
+      "grad_norm": 0.30682595841991545,
+      "learning_rate": 2e-05,
+      "loss": 0.0733,
+      "step": 275
+    },
+    {
+      "epoch": 2.3552,
+      "grad_norm": 0.2844580109974713,
+      "learning_rate": 2e-05,
+      "loss": 0.0979,
+      "step": 276
+    },
+    {
+      "epoch": 2.3637333333333332,
+      "grad_norm": 0.31296149985818705,
+      "learning_rate": 2e-05,
+      "loss": 0.093,
+      "step": 277
+    },
+    {
+      "epoch": 2.3722666666666665,
+      "grad_norm": 0.24448590086634314,
+      "learning_rate": 2e-05,
+      "loss": 0.1078,
+      "step": 278
+    },
+    {
+      "epoch": 2.3808,
+      "grad_norm": 0.17626685777153123,
+      "learning_rate": 2e-05,
+      "loss": 0.0939,
+      "step": 279
+    },
+    {
+      "epoch": 2.389333333333333,
+      "grad_norm": 0.13337416253457104,
+      "learning_rate": 2e-05,
+      "loss": 0.0812,
+      "step": 280
+    },
+    {
+      "epoch": 2.397866666666667,
+      "grad_norm": 0.13083267651254354,
+      "learning_rate": 2e-05,
+      "loss": 0.0929,
+      "step": 281
+    },
+    {
+      "epoch": 2.4064,
+      "grad_norm": 0.10994823562004175,
+      "learning_rate": 2e-05,
+      "loss": 0.098,
+      "step": 282
+    },
+    {
+      "epoch": 2.4149333333333334,
+      "grad_norm": 0.11578539664581784,
+      "learning_rate": 2e-05,
+      "loss": 0.098,
+      "step": 283
+    },
+    {
+      "epoch": 2.4234666666666667,
+      "grad_norm": 0.13153226988739827,
+      "learning_rate": 2e-05,
+      "loss": 0.0865,
+      "step": 284
+    },
+    {
+      "epoch": 2.432,
+      "grad_norm": 0.14019083181189823,
+      "learning_rate": 2e-05,
+      "loss": 0.1058,
+      "step": 285
+    },
+    {
+      "epoch": 2.440533333333333,
+      "grad_norm": 0.15458456814813418,
+      "learning_rate": 2e-05,
+      "loss": 0.0961,
+      "step": 286
+    },
+    {
+      "epoch": 2.4490666666666665,
+      "grad_norm": 0.13535285938076497,
+      "learning_rate": 2e-05,
+      "loss": 0.0919,
+      "step": 287
+    },
+    {
+      "epoch": 2.4576000000000002,
+      "grad_norm": 0.13511973312198838,
+      "learning_rate": 2e-05,
+      "loss": 0.0917,
+      "step": 288
+    },
+    {
+      "epoch": 2.4661333333333335,
+      "grad_norm": 0.12593494881253786,
+      "learning_rate": 2e-05,
+      "loss": 0.0894,
+      "step": 289
+    },
+    {
+      "epoch": 2.474666666666667,
+      "grad_norm": 0.11749403519277489,
+      "learning_rate": 2e-05,
+      "loss": 0.0701,
+      "step": 290
+    },
+    {
+      "epoch": 2.4832,
+      "grad_norm": 0.09252484293260567,
+      "learning_rate": 2e-05,
+      "loss": 0.1034,
+      "step": 291
+    },
+    {
+      "epoch": 2.4917333333333334,
+      "grad_norm": 0.12694234802279894,
+      "learning_rate": 2e-05,
+      "loss": 0.1053,
+      "step": 292
+    },
+    {
+      "epoch": 2.5002666666666666,
+      "grad_norm": 0.08086766935276264,
+      "learning_rate": 2e-05,
+      "loss": 0.0645,
+      "step": 293
+    },
+    {
+      "epoch": 2.5088,
+      "grad_norm": 0.0761948300276639,
+      "learning_rate": 2e-05,
+      "loss": 0.0825,
+      "step": 294
+    },
+    {
+      "epoch": 2.517333333333333,
+      "grad_norm": 0.05469199783697831,
+      "learning_rate": 2e-05,
+      "loss": 0.0809,
+      "step": 295
+    },
+    {
+      "epoch": 2.5258666666666665,
+      "grad_norm": 0.052391133906216227,
+      "learning_rate": 2e-05,
+      "loss": 0.0702,
+      "step": 296
+    },
+    {
+      "epoch": 2.5343999999999998,
+      "grad_norm": 0.05612973089592643,
+      "learning_rate": 2e-05,
+      "loss": 0.0739,
+      "step": 297
+    },
+    {
+      "epoch": 2.5429333333333335,
+      "grad_norm": 0.06163544903262162,
+      "learning_rate": 2e-05,
+      "loss": 0.0787,
+      "step": 298
+    },
+    {
+      "epoch": 2.5514666666666668,
+      "grad_norm": 0.04805683850668383,
+      "learning_rate": 2e-05,
+      "loss": 0.0781,
+      "step": 299
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.04677194701582553,
+      "learning_rate": 2e-05,
+      "loss": 0.0808,
+      "step": 300
+    }
+  ],
+  "logging_steps": 1,
+  "max_steps": 351,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 3,
+  "save_steps": 20,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 4.2831937920617677e+18,
+  "train_batch_size": 16,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-300/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aa4edad1da5557fcd52a6da980443588016cead6f0444a3562cfa68029c66a04
+size 6840

checkpoint-300/zero_to_fp32.py

@@ -0,0 +1,592 @@
+#!/usr/bin/env python
+
+# Copyright (c) Microsoft Corporation.
+# SPDX-License-Identifier: Apache-2.0
+
+# DeepSpeed Team
+
+# This script extracts fp32 consolidated weights from a zero 1, 2 and 3 DeepSpeed checkpoints. It gets
+# copied into the top level checkpoint dir, so the user can easily do the conversion at any point in
+# the future. Once extracted, the weights don't require DeepSpeed and can be used in any
+# application.
+#
+# example: python zero_to_fp32.py . pytorch_model.bin
+
+import argparse
+import torch
+import glob
+import math
+import os
+import re
+from collections import OrderedDict
+from dataclasses import dataclass
+
+# while this script doesn't use deepspeed to recover data, since the checkpoints are pickled with
+# DeepSpeed data structures it has to be available in the current python environment.
+from deepspeed.utils import logger
+from deepspeed.checkpoint.constants import (DS_VERSION, OPTIMIZER_STATE_DICT, SINGLE_PARTITION_OF_FP32_GROUPS,
+                                            FP32_FLAT_GROUPS, ZERO_STAGE, PARTITION_COUNT, PARAM_SHAPES, BUFFER_NAMES,
+                                            FROZEN_PARAM_SHAPES, FROZEN_PARAM_FRAGMENTS)
+
+
+@dataclass
+class zero_model_state:
+    buffers: dict()
+    param_shapes: dict()
+    shared_params: list
+    ds_version: int
+    frozen_param_shapes: dict()
+    frozen_param_fragments: dict()
+
+
+debug = 0
+
+# load to cpu
+device = torch.device('cpu')
+
+
+def atoi(text):
+    return int(text) if text.isdigit() else text
+
+
+def natural_keys(text):
+    '''
+    alist.sort(key=natural_keys) sorts in human order
+    http://nedbatchelder.com/blog/200712/human_sorting.html
+    (See Toothy's implementation in the comments)
+    '''
+    return [atoi(c) for c in re.split(r'(\d+)', text)]
+
+
+def get_model_state_file(checkpoint_dir, zero_stage):
+    if not os.path.isdir(checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{checkpoint_dir}' doesn't exist")
+
+    # there should be only one file
+    if zero_stage <= 2:
+        file = os.path.join(checkpoint_dir, "mp_rank_00_model_states.pt")
+    elif zero_stage == 3:
+        file = os.path.join(checkpoint_dir, "zero_pp_rank_0_mp_rank_00_model_states.pt")
+
+    if not os.path.exists(file):
+        raise FileNotFoundError(f"can't find model states file at '{file}'")
+
+    return file
+
+
+def get_checkpoint_files(checkpoint_dir, glob_pattern):
+    # XXX: need to test that this simple glob rule works for multi-node setup too
+    ckpt_files = sorted(glob.glob(os.path.join(checkpoint_dir, glob_pattern)), key=natural_keys)
+
+    if len(ckpt_files) == 0:
+        raise FileNotFoundError(f"can't find {glob_pattern} files in directory '{checkpoint_dir}'")
+
+    return ckpt_files
+
+
+def get_optim_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_optim_states.pt")
+
+
+def get_model_state_files(checkpoint_dir):
+    return get_checkpoint_files(checkpoint_dir, "*_model_states.pt")
+
+
+def parse_model_states(files):
+    zero_model_states = []
+    for file in files:
+        state_dict = torch.load(file, map_location=device)
+
+        if BUFFER_NAMES not in state_dict:
+            raise ValueError(f"{file} is not a model state checkpoint")
+        buffer_names = state_dict[BUFFER_NAMES]
+        if debug:
+            print("Found buffers:", buffer_names)
+
+        # recover just the buffers while restoring them to fp32 if they were saved in fp16
+        buffers = {k: v.float() for k, v in state_dict["module"].items() if k in buffer_names}
+        param_shapes = state_dict[PARAM_SHAPES]
+
+        # collect parameters that are included in param_shapes
+        param_names = []
+        for s in param_shapes:
+            for name in s.keys():
+                param_names.append(name)
+
+        # update with frozen parameters
+        frozen_param_shapes = state_dict.get(FROZEN_PARAM_SHAPES, None)
+        if frozen_param_shapes is not None:
+            if debug:
+                print(f"Found frozen_param_shapes: {frozen_param_shapes}")
+            param_names += list(frozen_param_shapes.keys())
+
+        # handle shared params
+        shared_params = [[k, v] for k, v in state_dict["shared_params"].items()]
+
+        ds_version = state_dict.get(DS_VERSION, None)
+
+        frozen_param_fragments = state_dict.get(FROZEN_PARAM_FRAGMENTS, None)
+
+        z_model_state = zero_model_state(buffers=buffers,
+                                         param_shapes=param_shapes,
+                                         shared_params=shared_params,
+                                         ds_version=ds_version,
+                                         frozen_param_shapes=frozen_param_shapes,
+                                         frozen_param_fragments=frozen_param_fragments)
+        zero_model_states.append(z_model_state)
+
+    return zero_model_states
+
+
+def parse_optim_states(files, ds_checkpoint_dir):
+
+    total_files = len(files)
+    state_dicts = []
+    for f in files:
+        state_dict = torch.load(f, map_location=device)
+        # immediately discard the potentially huge 2 optimizer states as we only care for fp32 master weights
+        # and also handle the case where it was already removed by another helper script
+        state_dict["optimizer_state_dict"].pop("optimizer_state_dict", None)
+        state_dicts.append(state_dict)
+
+    if not ZERO_STAGE in state_dicts[0][OPTIMIZER_STATE_DICT]:
+        raise ValueError(f"{files[0]} is not a zero checkpoint")
+    zero_stage = state_dicts[0][OPTIMIZER_STATE_DICT][ZERO_STAGE]
+    world_size = state_dicts[0][OPTIMIZER_STATE_DICT][PARTITION_COUNT]
+
+    # For ZeRO-2 each param group can have different partition_count as data parallelism for expert
+    # parameters can be different from data parallelism for non-expert parameters. So we can just
+    # use the max of the partition_count to get the dp world_size.
+
+    if type(world_size) is list:
+        world_size = max(world_size)
+
+    if world_size != total_files:
+        raise ValueError(
+            f"Expected {world_size} of '*_optim_states.pt' under '{ds_checkpoint_dir}' but found {total_files} files. "
+            "Possibly due to an overwrite of an old checkpoint, or a checkpoint didn't get saved by one or more processes."
+        )
+
+    # the groups are named differently in each stage
+    if zero_stage <= 2:
+        fp32_groups_key = SINGLE_PARTITION_OF_FP32_GROUPS
+    elif zero_stage == 3:
+        fp32_groups_key = FP32_FLAT_GROUPS
+    else:
+        raise ValueError(f"unknown zero stage {zero_stage}")
+
+    if zero_stage <= 2:
+        fp32_flat_groups = [state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key] for i in range(len(state_dicts))]
+    elif zero_stage == 3:
+        # if there is more than one param group, there will be multiple flattened tensors - one
+        # flattened tensor per group - for simplicity merge them into a single tensor
+        #
+        # XXX: could make the script more memory efficient for when there are multiple groups - it
+        # will require matching the sub-lists of param_shapes for each param group flattened tensor
+
+        fp32_flat_groups = [
+            torch.cat(state_dicts[i][OPTIMIZER_STATE_DICT][fp32_groups_key], 0) for i in range(len(state_dicts))
+        ]
+
+    return zero_stage, world_size, fp32_flat_groups
+
+
+def _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir):
+    """
+    Returns fp32 state_dict reconstructed from ds checkpoint
+
+    Args:
+        - ``ds_checkpoint_dir``: path to the deepspeed checkpoint folder (where the optimizer files are)
+
+    """
+    print(f"Processing zero checkpoint '{ds_checkpoint_dir}'")
+
+    optim_files = get_optim_files(ds_checkpoint_dir)
+    zero_stage, world_size, fp32_flat_groups = parse_optim_states(optim_files, ds_checkpoint_dir)
+    print(f"Detected checkpoint of type zero stage {zero_stage}, world_size: {world_size}")
+
+    model_files = get_model_state_files(ds_checkpoint_dir)
+
+    zero_model_states = parse_model_states(model_files)
+    print(f'Parsing checkpoint created by deepspeed=={zero_model_states[0].ds_version}')
+
+    if zero_stage <= 2:
+        return _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states)
+
+
+def _zero2_merge_frozen_params(state_dict, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+    frozen_param_fragments = zero_model_states[0].frozen_param_fragments
+
+    if debug:
+        num_elem = sum(s.numel() for s in frozen_param_shapes.values())
+        print(f'rank 0: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in frozen_param_fragments.values()])
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        state_dict[name] = frozen_param_fragments[name]
+
+        if debug:
+            print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _has_callable(obj, fn):
+    attr = getattr(obj, fn, None)
+    return callable(attr)
+
+
+def _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+
+    # Reconstruction protocol:
+    #
+    # XXX: document this
+
+    if debug:
+        for i in range(world_size):
+            for j in range(len(fp32_flat_groups[0])):
+                print(f"{FP32_FLAT_GROUPS}[{i}][{j}].shape={fp32_flat_groups[i][j].shape}")
+
+    # XXX: memory usage doubles here (zero2)
+    num_param_groups = len(fp32_flat_groups[0])
+    merged_single_partition_of_fp32_groups = []
+    for i in range(num_param_groups):
+        merged_partitions = [sd[i] for sd in fp32_flat_groups]
+        full_single_fp32_vector = torch.cat(merged_partitions, 0)
+        merged_single_partition_of_fp32_groups.append(full_single_fp32_vector)
+    avail_numel = sum(
+        [full_single_fp32_vector.numel() for full_single_fp32_vector in merged_single_partition_of_fp32_groups])
+
+    if debug:
+        wanted_params = sum([len(shapes) for shapes in param_shapes])
+        wanted_numel = sum([sum(shape.numel() for shape in shapes.values()) for shapes in param_shapes])
+        # not asserting if there is a mismatch due to possible padding
+        print(f"Have {avail_numel} numels to process.")
+        print(f"Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    total_numel = 0
+    total_params = 0
+    for shapes, full_single_fp32_vector in zip(param_shapes, merged_single_partition_of_fp32_groups):
+        offset = 0
+        avail_numel = full_single_fp32_vector.numel()
+        for name, shape in shapes.items():
+
+            unpartitioned_numel = shape.numel() if _has_callable(shape, 'numel') else math.prod(shape)
+            total_numel += unpartitioned_numel
+            total_params += 1
+
+            if debug:
+                print(f"{name} full shape: {shape} unpartitioned numel {unpartitioned_numel} ")
+            state_dict[name] = full_single_fp32_vector.narrow(0, offset, unpartitioned_numel).view(shape)
+            offset += unpartitioned_numel
+
+        # Z2 started to align to 2*world_size to improve nccl performance. Therefore both offset and
+        # avail_numel can differ by anywhere between 0..2*world_size. Due to two unrelated complex
+        # paddings performed in the code it's almost impossible to predict the exact numbers w/o the
+        # live optimizer object, so we are checking that the numbers are within the right range
+        align_to = 2 * world_size
+
+        def zero2_align(x):
+            return align_to * math.ceil(x / align_to)
+
+        if debug:
+            print(f"original offset={offset}, avail_numel={avail_numel}")
+
+        offset = zero2_align(offset)
+        avail_numel = zero2_align(avail_numel)
+
+        if debug:
+            print(f"aligned  offset={offset}, avail_numel={avail_numel}")
+
+        # Sanity check
+        if offset != avail_numel:
+            raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero2_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero2_merge_frozen_params(state_dict, zero_model_states)
+
+    _zero2_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def zero3_partitioned_param_info(unpartitioned_numel, world_size):
+    remainder = unpartitioned_numel % world_size
+    padding_numel = (world_size - remainder) if remainder else 0
+    partitioned_numel = math.ceil(unpartitioned_numel / world_size)
+    return partitioned_numel, padding_numel
+
+
+def _zero3_merge_frozen_params(state_dict, world_size, zero_model_states):
+    if zero_model_states[0].frozen_param_shapes is None or len(zero_model_states[0].frozen_param_shapes) == 0:
+        return
+
+    if debug:
+        for i in range(world_size):
+            num_elem = sum(s.numel() for s in zero_model_states[i].frozen_param_fragments.values())
+            print(f'rank {i}: {FROZEN_PARAM_SHAPES}.numel = {num_elem}')
+
+        frozen_param_shapes = zero_model_states[0].frozen_param_shapes
+        wanted_params = len(frozen_param_shapes)
+        wanted_numel = sum(s.numel() for s in frozen_param_shapes.values())
+        avail_numel = sum([p.numel() for p in zero_model_states[0].frozen_param_fragments.values()]) * world_size
+        print(f'Frozen params: Have {avail_numel} numels to process.')
+        print(f'Frozen params: Need {wanted_numel} numels in {wanted_params} params')
+
+    total_params = 0
+    total_numel = 0
+    for name, shape in zero_model_states[0].frozen_param_shapes.items():
+        total_params += 1
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+
+        param_frags = tuple(model_state.frozen_param_fragments[name] for model_state in zero_model_states)
+        state_dict[name] = torch.cat(param_frags, 0).narrow(0, 0, unpartitioned_numel).view(shape)
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Frozen params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+    print(f"Reconstructed Frozen fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states):
+    param_shapes = zero_model_states[0].param_shapes
+    avail_numel = fp32_flat_groups[0].numel() * world_size
+    # Reconstruction protocol: For zero3 we need to zip the partitions together at boundary of each
+    # param, re-consolidating each param, while dealing with padding if any
+
+    # merge list of dicts, preserving order
+    param_shapes = {k: v for d in param_shapes for k, v in d.items()}
+
+    if debug:
+        for i in range(world_size):
+            print(f"{FP32_FLAT_GROUPS}[{i}].shape={fp32_flat_groups[i].shape}")
+
+        wanted_params = len(param_shapes)
+        wanted_numel = sum(shape.numel() for shape in param_shapes.values())
+        # not asserting if there is a mismatch due to possible padding
+        avail_numel = fp32_flat_groups[0].numel() * world_size
+        print(f"Trainable params: Have {avail_numel} numels to process.")
+        print(f"Trainable params: Need {wanted_numel} numels in {wanted_params} params.")
+
+    # params
+    # XXX: for huge models that can't fit into the host's RAM we will have to recode this to support
+    # out-of-core computing solution
+    offset = 0
+    total_numel = 0
+    total_params = 0
+    for name, shape in param_shapes.items():
+
+        unpartitioned_numel = shape.numel()
+        total_numel += unpartitioned_numel
+        total_params += 1
+
+        partitioned_numel, partitioned_padding_numel = zero3_partitioned_param_info(unpartitioned_numel, world_size)
+
+        if debug:
+            print(
+                f"Trainable params: {total_params} {name} full shape: {shape} partition0 numel={partitioned_numel} partitioned_padding_numel={partitioned_padding_numel}"
+            )
+
+        # XXX: memory usage doubles here
+        state_dict[name] = torch.cat(
+            tuple(fp32_flat_groups[i].narrow(0, offset, partitioned_numel) for i in range(world_size)),
+            0).narrow(0, 0, unpartitioned_numel).view(shape)
+        offset += partitioned_numel
+
+    offset *= world_size
+
+    # Sanity check
+    if offset != avail_numel:
+        raise ValueError(f"consumed {offset} numels out of {avail_numel} - something is wrong")
+
+    print(f"Reconstructed Trainable fp32 state dict with {total_params} params {total_numel} elements")
+
+
+def _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    _zero3_merge_frozen_params(state_dict, world_size, zero_model_states)
+
+    _zero3_merge_trainable_params(state_dict, world_size, fp32_flat_groups, zero_model_states)
+
+    # recover shared parameters
+    for pair in zero_model_states[0].shared_params:
+        if pair[1] in state_dict:
+            state_dict[pair[0]] = state_dict[pair[1]]
+
+    return state_dict
+
+
+def get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated state_dict that can be loaded with
+    ``load_state_dict()`` and used for training without DeepSpeed or shared with others, for example
+    via a model hub.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in 'latest' file. e.g., ``global_step14``
+
+    Returns:
+        - pytorch ``state_dict``
+
+    Note: this approach may not work if your application doesn't have sufficient free CPU memory and
+    you may need to use the offline approach using the ``zero_to_fp32.py`` script that is saved with
+    the checkpoint.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import get_fp32_state_dict_from_zero_checkpoint
+        # do the training and checkpoint saving
+        state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir) # already on cpu
+        model = model.cpu() # move to cpu
+        model.load_state_dict(state_dict)
+        # submit to model hub or save the model to share with others
+
+    In this example the ``model`` will no longer be usable in the deepspeed context of the same
+    application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    If you want it all done for you, use ``load_state_dict_from_zero_checkpoint`` instead.
+
+    """
+    if tag is None:
+        latest_path = os.path.join(checkpoint_dir, 'latest')
+        if os.path.isfile(latest_path):
+            with open(latest_path, 'r') as fd:
+                tag = fd.read().strip()
+        else:
+            raise ValueError(f"Unable to find 'latest' file at {latest_path}")
+
+    ds_checkpoint_dir = os.path.join(checkpoint_dir, tag)
+
+    if not os.path.isdir(ds_checkpoint_dir):
+        raise FileNotFoundError(f"Directory '{ds_checkpoint_dir}' doesn't exist")
+
+    return _get_fp32_state_dict_from_zero_checkpoint(ds_checkpoint_dir)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None):
+    """
+    Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict`` file that can be
+    loaded with ``torch.load(file)`` + ``load_state_dict()`` and used for training without DeepSpeed.
+
+    Args:
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``output_file``: path to the pytorch fp32 state_dict output file (e.g. path/pytorch_model.bin)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+    print(f"Saving fp32 state dict to {output_file}")
+    torch.save(state_dict, output_file)
+
+
+def load_state_dict_from_zero_checkpoint(model, checkpoint_dir, tag=None):
+    """
+    1. Put the provided model to cpu
+    2. Convert ZeRO 2 or 3 checkpoint into a single fp32 consolidated ``state_dict``
+    3. Load it into the provided model
+
+    Args:
+        - ``model``: the model object to update
+        - ``checkpoint_dir``: path to the desired checkpoint folder. (one that contains the tag-folder, like ``global_step14``)
+        - ``tag``: checkpoint tag used as a unique identifier for checkpoint. If not provided will attempt to load tag in the file named ``latest`` in the checkpoint folder, e.g., ``global_step14``
+
+    Returns:
+        - ``model`: modified model
+
+    Make sure you have plenty of CPU memory available before you call this function. If you don't
+    have enough use the ``zero_to_fp32.py`` utility to do the conversion. You will find it
+    conveniently placed for you in the checkpoint folder.
+
+    A typical usage might be ::
+
+        from deepspeed.utils.zero_to_fp32 import load_state_dict_from_zero_checkpoint
+        model = load_state_dict_from_zero_checkpoint(trainer.model, checkpoint_dir)
+        # submit to model hub or save the model to share with others
+
+    Note, that once this was run, the ``model`` will no longer be usable in the deepspeed context
+    of the same application. i.e. you will need to re-initialize the deepspeed engine, since
+    ``model.load_state_dict(state_dict)`` will remove all the deepspeed magic from it.
+
+    """
+    logger.info(f"Extracting fp32 weights")
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag)
+
+    logger.info(f"Overwriting model with fp32 weights")
+    model = model.cpu()
+    model.load_state_dict(state_dict, strict=False)
+
+    return model
+
+
+if __name__ == "__main__":
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("checkpoint_dir",
+                        type=str,
+                        help="path to the desired checkpoint folder, e.g., path/checkpoint-12")
+    parser.add_argument(
+        "output_file",
+        type=str,
+        help="path to the pytorch fp32 state_dict output file (e.g. path/checkpoint-12/pytorch_model.bin)")
+    parser.add_argument("-t",
+                        "--tag",
+                        type=str,
+                        default=None,
+                        help="checkpoint tag used as a unique identifier for checkpoint. e.g., global_step1")
+    parser.add_argument("-d", "--debug", action='store_true', help="enable debug")
+    args = parser.parse_args()
+
+    debug = args.debug
+
+    convert_zero_checkpoint_to_fp32_state_dict(args.checkpoint_dir, args.output_file, tag=args.tag)