push llava-v1.6-mistral-7b

README.md

@@ -0,0 +1,202 @@
+---
+base_model: liuhaotian/llava-v1.6-mistral-7b
+library_name: peft
+---
+
+# 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 0.13.2

adapter_config.json

@@ -0,0 +1,34 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "liuhaotian/llava-v1.6-mistral-7b",
+  "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": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "q_proj",
+    "k_proj",
+    "down_proj",
+    "up_proj",
+    "gate_proj",
+    "v_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d1971af5a9b11de1f6838afa35ac3d724190147099272120bffd9b3c5163ce3
+size 44387440

checkpoint-125/README.md

@@ -0,0 +1,202 @@
+---
+base_model: liuhaotian/llava-v1.6-mistral-7b
+library_name: peft
+---
+
+# 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 0.13.2

checkpoint-125/adapter_config.json

@@ -0,0 +1,34 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "liuhaotian/llava-v1.6-mistral-7b",
+  "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": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "down_proj",
+    "q_proj",
+    "up_proj",
+    "o_proj",
+    "k_proj",
+    "v_proj",
+    "gate_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-125/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d1971af5a9b11de1f6838afa35ac3d724190147099272120bffd9b3c5163ce3
+size 44387440

checkpoint-125/global_step125/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3d49e00bc15726ea9dbea788cadae02446e9209f958323626568fe735627139f
+size 663858

checkpoint-125/global_step125/zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f4692ba939a6c56288bd8987774945c8129a1627131f57c31c6b6e0ee1f2fb18
+size 258789357

checkpoint-125/global_step125/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:066ba4e5576f9d8573646fedfcf414d3b9b1bbcaf8eea45f01baf46b6686bb8b
+size 663858

checkpoint-125/global_step125/zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7b9994ef4fede9b587d0236aa44b7b9720e1e1b5afa6aa5006c8e158b0406f26
+size 258789357

checkpoint-125/latest

@@ -0,0 +1 @@
+global_step125

checkpoint-125/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c0c87c21bd066bc64b06257b0cc8e31b0f640ffcf453027fd8e543ecab77a368
+size 14512

checkpoint-125/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ee81a002fb37d4cc0a9c2d509ae2a7b211b23ace5ecd3e43505f3a3110795529
+size 14512

checkpoint-125/special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-125/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
+size 493443

checkpoint-125/tokenizer_config.json

@@ -0,0 +1,45 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "chat_template": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 4096,
+  "pad_token": "<unk>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-125/trainer_state.json

@@ -0,0 +1,1908 @@
+{
+  "best_metric": 0.6666322350502014,
+  "best_model_checkpoint": "./checkpoints/llava-v1.6-mistral/checkpoint-125",
+  "epoch": 4.024,
+  "eval_steps": 1.0,
+  "global_step": 125,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.032,
+      "grad_norm": 1.3562940564009667,
+      "learning_rate": 0.0,
+      "loss": 1.2069,
+      "step": 1
+    },
+    {
+      "epoch": 0.032,
+      "eval_loss": 1.189530849456787,
+      "eval_runtime": 100.6384,
+      "eval_samples_per_second": 1.987,
+      "eval_steps_per_second": 0.497,
+      "step": 1
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 1.3726335882966116,
+      "learning_rate": 7.737056144690831e-06,
+      "loss": 1.242,
+      "step": 2
+    },
+    {
+      "epoch": 0.064,
+      "eval_loss": 1.189530849456787,
+      "eval_runtime": 95.3713,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 2
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 1.267071497789874,
+      "learning_rate": 1.2262943855309169e-05,
+      "loss": 1.1588,
+      "step": 3
+    },
+    {
+      "epoch": 0.096,
+      "eval_loss": 1.1675384044647217,
+      "eval_runtime": 99.4768,
+      "eval_samples_per_second": 2.011,
+      "eval_steps_per_second": 0.503,
+      "step": 3
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 1.2066781722006759,
+      "learning_rate": 1.5474112289381662e-05,
+      "loss": 1.1196,
+      "step": 4
+    },
+    {
+      "epoch": 0.128,
+      "eval_loss": 1.1391011476516724,
+      "eval_runtime": 95.4777,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 4
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2746210039866817,
+      "learning_rate": 1.7964888034078543e-05,
+      "loss": 1.1462,
+      "step": 5
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 1.1023523807525635,
+      "eval_runtime": 95.4138,
+      "eval_samples_per_second": 2.096,
+      "eval_steps_per_second": 0.524,
+      "step": 5
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 1.0552650848120237,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.1147,
+      "step": 6
+    },
+    {
+      "epoch": 0.192,
+      "eval_loss": 1.063745379447937,
+      "eval_runtime": 97.6633,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 6
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 1.0779287740182022,
+      "learning_rate": 2e-05,
+      "loss": 1.1773,
+      "step": 7
+    },
+    {
+      "epoch": 0.224,
+      "eval_loss": 1.024451732635498,
+      "eval_runtime": 97.388,
+      "eval_samples_per_second": 2.054,
+      "eval_steps_per_second": 0.513,
+      "step": 7
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 0.9918200495785524,
+      "learning_rate": 2e-05,
+      "loss": 1.0434,
+      "step": 8
+    },
+    {
+      "epoch": 0.256,
+      "eval_loss": 0.9906812906265259,
+      "eval_runtime": 97.2423,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 8
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 0.9119239308208719,
+      "learning_rate": 2e-05,
+      "loss": 1.0527,
+      "step": 9
+    },
+    {
+      "epoch": 0.288,
+      "eval_loss": 0.9626247882843018,
+      "eval_runtime": 97.2048,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.514,
+      "step": 9
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.7292073828818868,
+      "learning_rate": 2e-05,
+      "loss": 0.9801,
+      "step": 10
+    },
+    {
+      "epoch": 0.32,
+      "eval_loss": 0.9422003030776978,
+      "eval_runtime": 97.143,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 10
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 0.8195137858461904,
+      "learning_rate": 2e-05,
+      "loss": 1.0104,
+      "step": 11
+    },
+    {
+      "epoch": 0.352,
+      "eval_loss": 0.9264165759086609,
+      "eval_runtime": 97.2279,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 11
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 0.7060464565699565,
+      "learning_rate": 2e-05,
+      "loss": 0.9764,
+      "step": 12
+    },
+    {
+      "epoch": 0.384,
+      "eval_loss": 0.9138455986976624,
+      "eval_runtime": 97.0954,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 12
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 0.8207130544099034,
+      "learning_rate": 2e-05,
+      "loss": 1.0371,
+      "step": 13
+    },
+    {
+      "epoch": 0.416,
+      "eval_loss": 0.902658224105835,
+      "eval_runtime": 100.8924,
+      "eval_samples_per_second": 1.982,
+      "eval_steps_per_second": 0.496,
+      "step": 13
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 0.7962544586919155,
+      "learning_rate": 2e-05,
+      "loss": 0.9455,
+      "step": 14
+    },
+    {
+      "epoch": 0.448,
+      "eval_loss": 0.8926536440849304,
+      "eval_runtime": 100.856,
+      "eval_samples_per_second": 1.983,
+      "eval_steps_per_second": 0.496,
+      "step": 14
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.8246421268957655,
+      "learning_rate": 2e-05,
+      "loss": 0.9456,
+      "step": 15
+    },
+    {
+      "epoch": 0.48,
+      "eval_loss": 0.8823295831680298,
+      "eval_runtime": 97.0568,
+      "eval_samples_per_second": 2.061,
+      "eval_steps_per_second": 0.515,
+      "step": 15
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.7040845257818263,
+      "learning_rate": 2e-05,
+      "loss": 0.9283,
+      "step": 16
+    },
+    {
+      "epoch": 0.512,
+      "eval_loss": 0.8721897602081299,
+      "eval_runtime": 98.318,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 16
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 0.7904707927869743,
+      "learning_rate": 2e-05,
+      "loss": 0.9786,
+      "step": 17
+    },
+    {
+      "epoch": 0.544,
+      "eval_loss": 0.861838698387146,
+      "eval_runtime": 101.4487,
+      "eval_samples_per_second": 1.971,
+      "eval_steps_per_second": 0.493,
+      "step": 17
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 0.7605583546638608,
+      "learning_rate": 2e-05,
+      "loss": 0.9361,
+      "step": 18
+    },
+    {
+      "epoch": 0.576,
+      "eval_loss": 0.8519415259361267,
+      "eval_runtime": 98.2448,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 18
+    },
+    {
+      "epoch": 0.608,
+      "grad_norm": 0.7005232039249588,
+      "learning_rate": 2e-05,
+      "loss": 0.8707,
+      "step": 19
+    },
+    {
+      "epoch": 0.608,
+      "eval_loss": 0.842995822429657,
+      "eval_runtime": 98.3165,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 19
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.653693259495668,
+      "learning_rate": 2e-05,
+      "loss": 0.9332,
+      "step": 20
+    },
+    {
+      "epoch": 0.64,
+      "eval_loss": 0.8355565071105957,
+      "eval_runtime": 98.237,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 20
+    },
+    {
+      "epoch": 0.672,
+      "grad_norm": 0.6600451404133434,
+      "learning_rate": 2e-05,
+      "loss": 0.8848,
+      "step": 21
+    },
+    {
+      "epoch": 0.672,
+      "eval_loss": 0.8297473788261414,
+      "eval_runtime": 98.1825,
+      "eval_samples_per_second": 2.037,
+      "eval_steps_per_second": 0.509,
+      "step": 21
+    },
+    {
+      "epoch": 0.704,
+      "grad_norm": 0.6666525650519819,
+      "learning_rate": 2e-05,
+      "loss": 0.9261,
+      "step": 22
+    },
+    {
+      "epoch": 0.704,
+      "eval_loss": 0.8248497247695923,
+      "eval_runtime": 98.1541,
+      "eval_samples_per_second": 2.038,
+      "eval_steps_per_second": 0.509,
+      "step": 22
+    },
+    {
+      "epoch": 0.736,
+      "grad_norm": 0.6759718697778233,
+      "learning_rate": 2e-05,
+      "loss": 1.0103,
+      "step": 23
+    },
+    {
+      "epoch": 0.736,
+      "eval_loss": 0.8214980959892273,
+      "eval_runtime": 98.2494,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 23
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.651870418762904,
+      "learning_rate": 2e-05,
+      "loss": 0.9538,
+      "step": 24
+    },
+    {
+      "epoch": 0.768,
+      "eval_loss": 0.818403959274292,
+      "eval_runtime": 97.9476,
+      "eval_samples_per_second": 2.042,
+      "eval_steps_per_second": 0.51,
+      "step": 24
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.6985872283132858,
+      "learning_rate": 2e-05,
+      "loss": 0.8216,
+      "step": 25
+    },
+    {
+      "epoch": 0.8,
+      "eval_loss": 0.8149588108062744,
+      "eval_runtime": 95.6175,
+      "eval_samples_per_second": 2.092,
+      "eval_steps_per_second": 0.523,
+      "step": 25
+    },
+    {
+      "epoch": 0.832,
+      "grad_norm": 0.7801182369335715,
+      "learning_rate": 2e-05,
+      "loss": 0.8478,
+      "step": 26
+    },
+    {
+      "epoch": 0.832,
+      "eval_loss": 0.8100517392158508,
+      "eval_runtime": 99.7733,
+      "eval_samples_per_second": 2.005,
+      "eval_steps_per_second": 0.501,
+      "step": 26
+    },
+    {
+      "epoch": 0.864,
+      "grad_norm": 0.6727766374526198,
+      "learning_rate": 2e-05,
+      "loss": 0.9015,
+      "step": 27
+    },
+    {
+      "epoch": 0.864,
+      "eval_loss": 0.8051578998565674,
+      "eval_runtime": 95.7387,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 27
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.7398381040653764,
+      "learning_rate": 2e-05,
+      "loss": 0.8434,
+      "step": 28
+    },
+    {
+      "epoch": 0.896,
+      "eval_loss": 0.7998965382575989,
+      "eval_runtime": 95.7753,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 28
+    },
+    {
+      "epoch": 0.928,
+      "grad_norm": 0.6837341164482282,
+      "learning_rate": 2e-05,
+      "loss": 0.9124,
+      "step": 29
+    },
+    {
+      "epoch": 0.928,
+      "eval_loss": 0.7946869134902954,
+      "eval_runtime": 95.933,
+      "eval_samples_per_second": 2.085,
+      "eval_steps_per_second": 0.521,
+      "step": 29
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.7285508052728015,
+      "learning_rate": 2e-05,
+      "loss": 0.8252,
+      "step": 30
+    },
+    {
+      "epoch": 0.96,
+      "eval_loss": 0.7891057729721069,
+      "eval_runtime": 95.871,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.522,
+      "step": 30
+    },
+    {
+      "epoch": 0.992,
+      "grad_norm": 0.706524788728722,
+      "learning_rate": 2e-05,
+      "loss": 0.8382,
+      "step": 31
+    },
+    {
+      "epoch": 0.992,
+      "eval_loss": 0.7835636138916016,
+      "eval_runtime": 95.7491,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 31
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.7392763969473979,
+      "learning_rate": 2e-05,
+      "loss": 0.8586,
+      "step": 32
+    },
+    {
+      "epoch": 1.024,
+      "eval_loss": 0.7780101299285889,
+      "eval_runtime": 99.7493,
+      "eval_samples_per_second": 2.005,
+      "eval_steps_per_second": 0.501,
+      "step": 32
+    },
+    {
+      "epoch": 1.056,
+      "grad_norm": 0.7407143277408553,
+      "learning_rate": 2e-05,
+      "loss": 0.8438,
+      "step": 33
+    },
+    {
+      "epoch": 1.056,
+      "eval_loss": 0.7728690505027771,
+      "eval_runtime": 95.8863,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.521,
+      "step": 33
+    },
+    {
+      "epoch": 1.088,
+      "grad_norm": 0.7146296362894146,
+      "learning_rate": 2e-05,
+      "loss": 0.8357,
+      "step": 34
+    },
+    {
+      "epoch": 1.088,
+      "eval_loss": 0.7683935761451721,
+      "eval_runtime": 97.824,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 34
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.7117142678212836,
+      "learning_rate": 2e-05,
+      "loss": 0.7884,
+      "step": 35
+    },
+    {
+      "epoch": 1.12,
+      "eval_loss": 0.7646369338035583,
+      "eval_runtime": 95.8185,
+      "eval_samples_per_second": 2.087,
+      "eval_steps_per_second": 0.522,
+      "step": 35
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.7735767817076163,
+      "learning_rate": 2e-05,
+      "loss": 0.8851,
+      "step": 36
+    },
+    {
+      "epoch": 1.152,
+      "eval_loss": 0.7616337537765503,
+      "eval_runtime": 96.1525,
+      "eval_samples_per_second": 2.08,
+      "eval_steps_per_second": 0.52,
+      "step": 36
+    },
+    {
+      "epoch": 1.184,
+      "grad_norm": 0.7386952203362822,
+      "learning_rate": 2e-05,
+      "loss": 0.7943,
+      "step": 37
+    },
+    {
+      "epoch": 1.184,
+      "eval_loss": 0.759408712387085,
+      "eval_runtime": 96.2772,
+      "eval_samples_per_second": 2.077,
+      "eval_steps_per_second": 0.519,
+      "step": 37
+    },
+    {
+      "epoch": 1.216,
+      "grad_norm": 0.6688305832985106,
+      "learning_rate": 2e-05,
+      "loss": 0.8433,
+      "step": 38
+    },
+    {
+      "epoch": 1.216,
+      "eval_loss": 0.7577520608901978,
+      "eval_runtime": 95.7726,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 38
+    },
+    {
+      "epoch": 1.248,
+      "grad_norm": 0.6861808214947546,
+      "learning_rate": 2e-05,
+      "loss": 0.8132,
+      "step": 39
+    },
+    {
+      "epoch": 1.248,
+      "eval_loss": 0.7565059065818787,
+      "eval_runtime": 96.1577,
+      "eval_samples_per_second": 2.08,
+      "eval_steps_per_second": 0.52,
+      "step": 39
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.7450886292190618,
+      "learning_rate": 2e-05,
+      "loss": 0.9067,
+      "step": 40
+    },
+    {
+      "epoch": 1.28,
+      "eval_loss": 0.7562046647071838,
+      "eval_runtime": 95.9983,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 40
+    },
+    {
+      "epoch": 1.312,
+      "grad_norm": 0.7508586498301053,
+      "learning_rate": 2e-05,
+      "loss": 0.8642,
+      "step": 41
+    },
+    {
+      "epoch": 1.312,
+      "eval_loss": 0.755351722240448,
+      "eval_runtime": 96.6244,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 41
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "grad_norm": 0.7717273184776977,
+      "learning_rate": 2e-05,
+      "loss": 0.9406,
+      "step": 42
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "eval_loss": 0.753773033618927,
+      "eval_runtime": 100.8127,
+      "eval_samples_per_second": 1.984,
+      "eval_steps_per_second": 0.496,
+      "step": 42
+    },
+    {
+      "epoch": 1.376,
+      "grad_norm": 0.7980115302909253,
+      "learning_rate": 2e-05,
+      "loss": 0.7732,
+      "step": 43
+    },
+    {
+      "epoch": 1.376,
+      "eval_loss": 0.7511720061302185,
+      "eval_runtime": 97.4372,
+      "eval_samples_per_second": 2.053,
+      "eval_steps_per_second": 0.513,
+      "step": 43
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.794554904521861,
+      "learning_rate": 2e-05,
+      "loss": 0.8962,
+      "step": 44
+    },
+    {
+      "epoch": 1.408,
+      "eval_loss": 0.7484354376792908,
+      "eval_runtime": 96.7621,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 44
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.7964935137573818,
+      "learning_rate": 2e-05,
+      "loss": 0.7725,
+      "step": 45
+    },
+    {
+      "epoch": 1.44,
+      "eval_loss": 0.7454251050949097,
+      "eval_runtime": 96.5292,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 45
+    },
+    {
+      "epoch": 1.472,
+      "grad_norm": 0.7477437094752549,
+      "learning_rate": 2e-05,
+      "loss": 0.8238,
+      "step": 46
+    },
+    {
+      "epoch": 1.472,
+      "eval_loss": 0.7427694201469421,
+      "eval_runtime": 96.4788,
+      "eval_samples_per_second": 2.073,
+      "eval_steps_per_second": 0.518,
+      "step": 46
+    },
+    {
+      "epoch": 1.504,
+      "grad_norm": 0.7443339980176984,
+      "learning_rate": 2e-05,
+      "loss": 0.8273,
+      "step": 47
+    },
+    {
+      "epoch": 1.504,
+      "eval_loss": 0.7407447099685669,
+      "eval_runtime": 96.5686,
+      "eval_samples_per_second": 2.071,
+      "eval_steps_per_second": 0.518,
+      "step": 47
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.7901404920428264,
+      "learning_rate": 2e-05,
+      "loss": 0.7974,
+      "step": 48
+    },
+    {
+      "epoch": 1.536,
+      "eval_loss": 0.7381842136383057,
+      "eval_runtime": 96.6071,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.518,
+      "step": 48
+    },
+    {
+      "epoch": 1.568,
+      "grad_norm": 0.8677109521747711,
+      "learning_rate": 2e-05,
+      "loss": 0.8011,
+      "step": 49
+    },
+    {
+      "epoch": 1.568,
+      "eval_loss": 0.73555588722229,
+      "eval_runtime": 96.4189,
+      "eval_samples_per_second": 2.074,
+      "eval_steps_per_second": 0.519,
+      "step": 49
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.8170436579363425,
+      "learning_rate": 2e-05,
+      "loss": 0.7762,
+      "step": 50
+    },
+    {
+      "epoch": 1.6,
+      "eval_loss": 0.7331439256668091,
+      "eval_runtime": 100.5875,
+      "eval_samples_per_second": 1.988,
+      "eval_steps_per_second": 0.497,
+      "step": 50
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "grad_norm": 0.8190539924095326,
+      "learning_rate": 2e-05,
+      "loss": 0.7469,
+      "step": 51
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "eval_loss": 0.7306647300720215,
+      "eval_runtime": 101.1429,
+      "eval_samples_per_second": 1.977,
+      "eval_steps_per_second": 0.494,
+      "step": 51
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.8044932603192197,
+      "learning_rate": 2e-05,
+      "loss": 0.8052,
+      "step": 52
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "eval_loss": 0.7287429571151733,
+      "eval_runtime": 97.25,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 52
+    },
+    {
+      "epoch": 1.696,
+      "grad_norm": 0.7811170846407103,
+      "learning_rate": 2e-05,
+      "loss": 0.755,
+      "step": 53
+    },
+    {
+      "epoch": 1.696,
+      "eval_loss": 0.7264651656150818,
+      "eval_runtime": 97.67,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 53
+    },
+    {
+      "epoch": 1.728,
+      "grad_norm": 0.808405093976468,
+      "learning_rate": 2e-05,
+      "loss": 0.7838,
+      "step": 54
+    },
+    {
+      "epoch": 1.728,
+      "eval_loss": 0.7242828607559204,
+      "eval_runtime": 101.2008,
+      "eval_samples_per_second": 1.976,
+      "eval_steps_per_second": 0.494,
+      "step": 54
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.8172106270954758,
+      "learning_rate": 2e-05,
+      "loss": 0.7123,
+      "step": 55
+    },
+    {
+      "epoch": 1.76,
+      "eval_loss": 0.7217574119567871,
+      "eval_runtime": 99.6145,
+      "eval_samples_per_second": 2.008,
+      "eval_steps_per_second": 0.502,
+      "step": 55
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.83212530473105,
+      "learning_rate": 2e-05,
+      "loss": 0.8141,
+      "step": 56
+    },
+    {
+      "epoch": 1.792,
+      "eval_loss": 0.7192932367324829,
+      "eval_runtime": 97.1377,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 56
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "grad_norm": 0.8632612959683078,
+      "learning_rate": 2e-05,
+      "loss": 0.8005,
+      "step": 57
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "eval_loss": 0.717003345489502,
+      "eval_runtime": 97.2871,
+      "eval_samples_per_second": 2.056,
+      "eval_steps_per_second": 0.514,
+      "step": 57
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "grad_norm": 0.8362081570841255,
+      "learning_rate": 2e-05,
+      "loss": 0.8254,
+      "step": 58
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "eval_loss": 0.715446949005127,
+      "eval_runtime": 97.5751,
+      "eval_samples_per_second": 2.05,
+      "eval_steps_per_second": 0.512,
+      "step": 58
+    },
+    {
+      "epoch": 1.888,
+      "grad_norm": 0.8862779841020042,
+      "learning_rate": 2e-05,
+      "loss": 0.7949,
+      "step": 59
+    },
+    {
+      "epoch": 1.888,
+      "eval_loss": 0.7145370244979858,
+      "eval_runtime": 97.2406,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 59
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.875656256488838,
+      "learning_rate": 2e-05,
+      "loss": 0.8277,
+      "step": 60
+    },
+    {
+      "epoch": 1.92,
+      "eval_loss": 0.7140547037124634,
+      "eval_runtime": 97.1485,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 60
+    },
+    {
+      "epoch": 1.952,
+      "grad_norm": 0.8850690645543244,
+      "learning_rate": 2e-05,
+      "loss": 0.8117,
+      "step": 61
+    },
+    {
+      "epoch": 1.952,
+      "eval_loss": 0.7128701210021973,
+      "eval_runtime": 97.8906,
+      "eval_samples_per_second": 2.043,
+      "eval_steps_per_second": 0.511,
+      "step": 61
+    },
+    {
+      "epoch": 1.984,
+      "grad_norm": 0.8961641824408927,
+      "learning_rate": 2e-05,
+      "loss": 0.704,
+      "step": 62
+    },
+    {
+      "epoch": 1.984,
+      "eval_loss": 0.7107064127922058,
+      "eval_runtime": 102.2163,
+      "eval_samples_per_second": 1.957,
+      "eval_steps_per_second": 0.489,
+      "step": 62
+    },
+    {
+      "epoch": 2.016,
+      "grad_norm": 0.8602247372004996,
+      "learning_rate": 2e-05,
+      "loss": 0.8016,
+      "step": 63
+    },
+    {
+      "epoch": 2.016,
+      "eval_loss": 0.7091581225395203,
+      "eval_runtime": 98.2114,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 63
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.7881703384753784,
+      "learning_rate": 2e-05,
+      "loss": 0.7912,
+      "step": 64
+    },
+    {
+      "epoch": 2.048,
+      "eval_loss": 0.7081906199455261,
+      "eval_runtime": 98.9403,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 64
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.8436680506716614,
+      "learning_rate": 2e-05,
+      "loss": 0.6965,
+      "step": 65
+    },
+    {
+      "epoch": 2.08,
+      "eval_loss": 0.7070262432098389,
+      "eval_runtime": 97.7451,
+      "eval_samples_per_second": 2.046,
+      "eval_steps_per_second": 0.512,
+      "step": 65
+    },
+    {
+      "epoch": 2.112,
+      "grad_norm": 0.8694446846234115,
+      "learning_rate": 2e-05,
+      "loss": 0.7015,
+      "step": 66
+    },
+    {
+      "epoch": 2.112,
+      "eval_loss": 0.7055197358131409,
+      "eval_runtime": 97.994,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 66
+    },
+    {
+      "epoch": 2.144,
+      "grad_norm": 0.9518475022730614,
+      "learning_rate": 2e-05,
+      "loss": 0.6952,
+      "step": 67
+    },
+    {
+      "epoch": 2.144,
+      "eval_loss": 0.7035844326019287,
+      "eval_runtime": 98.0296,
+      "eval_samples_per_second": 2.04,
+      "eval_steps_per_second": 0.51,
+      "step": 67
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.8662066024104106,
+      "learning_rate": 2e-05,
+      "loss": 0.7731,
+      "step": 68
+    },
+    {
+      "epoch": 2.176,
+      "eval_loss": 0.7019283771514893,
+      "eval_runtime": 100.0982,
+      "eval_samples_per_second": 1.998,
+      "eval_steps_per_second": 0.5,
+      "step": 68
+    },
+    {
+      "epoch": 2.208,
+      "grad_norm": 0.9574631942368209,
+      "learning_rate": 2e-05,
+      "loss": 0.7572,
+      "step": 69
+    },
+    {
+      "epoch": 2.208,
+      "eval_loss": 0.7011401653289795,
+      "eval_runtime": 101.9267,
+      "eval_samples_per_second": 1.962,
+      "eval_steps_per_second": 0.491,
+      "step": 69
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.9477542521472675,
+      "learning_rate": 2e-05,
+      "loss": 0.7393,
+      "step": 70
+    },
+    {
+      "epoch": 2.24,
+      "eval_loss": 0.7006180286407471,
+      "eval_runtime": 101.9672,
+      "eval_samples_per_second": 1.961,
+      "eval_steps_per_second": 0.49,
+      "step": 70
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "grad_norm": 0.9612956116350272,
+      "learning_rate": 2e-05,
+      "loss": 0.7175,
+      "step": 71
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "eval_loss": 0.7002539038658142,
+      "eval_runtime": 101.8755,
+      "eval_samples_per_second": 1.963,
+      "eval_steps_per_second": 0.491,
+      "step": 71
+    },
+    {
+      "epoch": 2.304,
+      "grad_norm": 0.9934844303955727,
+      "learning_rate": 2e-05,
+      "loss": 0.791,
+      "step": 72
+    },
+    {
+      "epoch": 2.304,
+      "eval_loss": 0.6997203230857849,
+      "eval_runtime": 97.9926,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 72
+    },
+    {
+      "epoch": 2.336,
+      "grad_norm": 1.0540430431227044,
+      "learning_rate": 2e-05,
+      "loss": 0.7542,
+      "step": 73
+    },
+    {
+      "epoch": 2.336,
+      "eval_loss": 0.6988361477851868,
+      "eval_runtime": 100.3704,
+      "eval_samples_per_second": 1.993,
+      "eval_steps_per_second": 0.498,
+      "step": 73
+    },
+    {
+      "epoch": 2.368,
+      "grad_norm": 1.0249397957961794,
+      "learning_rate": 2e-05,
+      "loss": 0.8084,
+      "step": 74
+    },
+    {
+      "epoch": 2.368,
+      "eval_loss": 0.6980065703392029,
+      "eval_runtime": 101.8958,
+      "eval_samples_per_second": 1.963,
+      "eval_steps_per_second": 0.491,
+      "step": 74
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.0445498365690145,
+      "learning_rate": 2e-05,
+      "loss": 0.7964,
+      "step": 75
+    },
+    {
+      "epoch": 2.4,
+      "eval_loss": 0.6971798539161682,
+      "eval_runtime": 98.1624,
+      "eval_samples_per_second": 2.037,
+      "eval_steps_per_second": 0.509,
+      "step": 75
+    },
+    {
+      "epoch": 2.432,
+      "grad_norm": 0.9685893079320761,
+      "learning_rate": 2e-05,
+      "loss": 0.8047,
+      "step": 76
+    },
+    {
+      "epoch": 2.432,
+      "eval_loss": 0.696861743927002,
+      "eval_runtime": 98.6404,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 76
+    },
+    {
+      "epoch": 2.464,
+      "grad_norm": 0.9753091933204456,
+      "learning_rate": 2e-05,
+      "loss": 0.684,
+      "step": 77
+    },
+    {
+      "epoch": 2.464,
+      "eval_loss": 0.6957904100418091,
+      "eval_runtime": 98.8766,
+      "eval_samples_per_second": 2.023,
+      "eval_steps_per_second": 0.506,
+      "step": 77
+    },
+    {
+      "epoch": 2.496,
+      "grad_norm": 0.9389149478503764,
+      "learning_rate": 2e-05,
+      "loss": 0.7434,
+      "step": 78
+    },
+    {
+      "epoch": 2.496,
+      "eval_loss": 0.6943306922912598,
+      "eval_runtime": 98.5751,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 78
+    },
+    {
+      "epoch": 2.528,
+      "grad_norm": 1.0675154845211299,
+      "learning_rate": 2e-05,
+      "loss": 0.7208,
+      "step": 79
+    },
+    {
+      "epoch": 2.528,
+      "eval_loss": 0.6920651197433472,
+      "eval_runtime": 99.0851,
+      "eval_samples_per_second": 2.018,
+      "eval_steps_per_second": 0.505,
+      "step": 79
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.9937936593307737,
+      "learning_rate": 2e-05,
+      "loss": 0.6948,
+      "step": 80
+    },
+    {
+      "epoch": 2.56,
+      "eval_loss": 0.6899142265319824,
+      "eval_runtime": 98.7974,
+      "eval_samples_per_second": 2.024,
+      "eval_steps_per_second": 0.506,
+      "step": 80
+    },
+    {
+      "epoch": 2.592,
+      "grad_norm": 0.9650832276698476,
+      "learning_rate": 2e-05,
+      "loss": 0.7666,
+      "step": 81
+    },
+    {
+      "epoch": 2.592,
+      "eval_loss": 0.6886695623397827,
+      "eval_runtime": 98.6813,
+      "eval_samples_per_second": 2.027,
+      "eval_steps_per_second": 0.507,
+      "step": 81
+    },
+    {
+      "epoch": 2.624,
+      "grad_norm": 0.9961610958296112,
+      "learning_rate": 2e-05,
+      "loss": 0.7741,
+      "step": 82
+    },
+    {
+      "epoch": 2.624,
+      "eval_loss": 0.687745213508606,
+      "eval_runtime": 98.624,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 82
+    },
+    {
+      "epoch": 2.656,
+      "grad_norm": 1.0261499699526089,
+      "learning_rate": 2e-05,
+      "loss": 0.7747,
+      "step": 83
+    },
+    {
+      "epoch": 2.656,
+      "eval_loss": 0.6869972944259644,
+      "eval_runtime": 98.9604,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 83
+    },
+    {
+      "epoch": 2.6879999999999997,
+      "grad_norm": 1.025059237745532,
+      "learning_rate": 2e-05,
+      "loss": 0.7526,
+      "step": 84
+    },
+    {
+      "epoch": 2.6879999999999997,
+      "eval_loss": 0.6862147450447083,
+      "eval_runtime": 98.4387,
+      "eval_samples_per_second": 2.032,
+      "eval_steps_per_second": 0.508,
+      "step": 84
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "grad_norm": 1.1383626266013125,
+      "learning_rate": 2e-05,
+      "loss": 0.683,
+      "step": 85
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_loss": 0.6845572590827942,
+      "eval_runtime": 98.4979,
+      "eval_samples_per_second": 2.03,
+      "eval_steps_per_second": 0.508,
+      "step": 85
+    },
+    {
+      "epoch": 2.752,
+      "grad_norm": 1.0427739679421295,
+      "learning_rate": 2e-05,
+      "loss": 0.7362,
+      "step": 86
+    },
+    {
+      "epoch": 2.752,
+      "eval_loss": 0.683280348777771,
+      "eval_runtime": 103.6707,
+      "eval_samples_per_second": 1.929,
+      "eval_steps_per_second": 0.482,
+      "step": 86
+    },
+    {
+      "epoch": 2.784,
+      "grad_norm": 1.1280129141879938,
+      "learning_rate": 2e-05,
+      "loss": 0.7743,
+      "step": 87
+    },
+    {
+      "epoch": 2.784,
+      "eval_loss": 0.6824235320091248,
+      "eval_runtime": 99.2995,
+      "eval_samples_per_second": 2.014,
+      "eval_steps_per_second": 0.504,
+      "step": 87
+    },
+    {
+      "epoch": 2.816,
+      "grad_norm": 1.0017715497784696,
+      "learning_rate": 2e-05,
+      "loss": 0.7164,
+      "step": 88
+    },
+    {
+      "epoch": 2.816,
+      "eval_loss": 0.6819013357162476,
+      "eval_runtime": 99.5206,
+      "eval_samples_per_second": 2.01,
+      "eval_steps_per_second": 0.502,
+      "step": 88
+    },
+    {
+      "epoch": 2.848,
+      "grad_norm": 1.0769673642284994,
+      "learning_rate": 2e-05,
+      "loss": 0.7013,
+      "step": 89
+    },
+    {
+      "epoch": 2.848,
+      "eval_loss": 0.6820746064186096,
+      "eval_runtime": 96.5197,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 89
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 1.0389167281844591,
+      "learning_rate": 2e-05,
+      "loss": 0.6805,
+      "step": 90
+    },
+    {
+      "epoch": 2.88,
+      "eval_loss": 0.6826525926589966,
+      "eval_runtime": 96.0942,
+      "eval_samples_per_second": 2.081,
+      "eval_steps_per_second": 0.52,
+      "step": 90
+    },
+    {
+      "epoch": 2.912,
+      "grad_norm": 0.9705743838620626,
+      "learning_rate": 2e-05,
+      "loss": 0.7681,
+      "step": 91
+    },
+    {
+      "epoch": 2.912,
+      "eval_loss": 0.6836435198783875,
+      "eval_runtime": 96.2441,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 91
+    },
+    {
+      "epoch": 2.944,
+      "grad_norm": 1.037823791993831,
+      "learning_rate": 2e-05,
+      "loss": 0.7532,
+      "step": 92
+    },
+    {
+      "epoch": 2.944,
+      "eval_loss": 0.6845746040344238,
+      "eval_runtime": 96.2394,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 92
+    },
+    {
+      "epoch": 2.976,
+      "grad_norm": 1.1323835942146157,
+      "learning_rate": 2e-05,
+      "loss": 0.7171,
+      "step": 93
+    },
+    {
+      "epoch": 2.976,
+      "eval_loss": 0.684663712978363,
+      "eval_runtime": 96.23,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 93
+    },
+    {
+      "epoch": 3.008,
+      "grad_norm": 1.1957864756602699,
+      "learning_rate": 2e-05,
+      "loss": 0.7166,
+      "step": 94
+    },
+    {
+      "epoch": 3.008,
+      "eval_loss": 0.6830846667289734,
+      "eval_runtime": 96.6549,
+      "eval_samples_per_second": 2.069,
+      "eval_steps_per_second": 0.517,
+      "step": 94
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 1.1077357794232636,
+      "learning_rate": 2e-05,
+      "loss": 0.6667,
+      "step": 95
+    },
+    {
+      "epoch": 3.04,
+      "eval_loss": 0.6810076832771301,
+      "eval_runtime": 96.3239,
+      "eval_samples_per_second": 2.076,
+      "eval_steps_per_second": 0.519,
+      "step": 95
+    },
+    {
+      "epoch": 3.072,
+      "grad_norm": 1.1851219157184936,
+      "learning_rate": 2e-05,
+      "loss": 0.7115,
+      "step": 96
+    },
+    {
+      "epoch": 3.072,
+      "eval_loss": 0.6796395778656006,
+      "eval_runtime": 96.9109,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 96
+    },
+    {
+      "epoch": 3.104,
+      "grad_norm": 1.0812671042616444,
+      "learning_rate": 2e-05,
+      "loss": 0.7333,
+      "step": 97
+    },
+    {
+      "epoch": 3.104,
+      "eval_loss": 0.6794567108154297,
+      "eval_runtime": 96.7403,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 97
+    },
+    {
+      "epoch": 3.136,
+      "grad_norm": 1.130095597839828,
+      "learning_rate": 2e-05,
+      "loss": 0.6328,
+      "step": 98
+    },
+    {
+      "epoch": 3.136,
+      "eval_loss": 0.6792007684707642,
+      "eval_runtime": 96.8136,
+      "eval_samples_per_second": 2.066,
+      "eval_steps_per_second": 0.516,
+      "step": 98
+    },
+    {
+      "epoch": 3.168,
+      "grad_norm": 1.16102100344116,
+      "learning_rate": 2e-05,
+      "loss": 0.6625,
+      "step": 99
+    },
+    {
+      "epoch": 3.168,
+      "eval_loss": 0.6789132952690125,
+      "eval_runtime": 96.6982,
+      "eval_samples_per_second": 2.068,
+      "eval_steps_per_second": 0.517,
+      "step": 99
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 1.226689811951201,
+      "learning_rate": 2e-05,
+      "loss": 0.7522,
+      "step": 100
+    },
+    {
+      "epoch": 3.2,
+      "eval_loss": 0.6786602735519409,
+      "eval_runtime": 96.6253,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 100
+    },
+    {
+      "epoch": 3.232,
+      "grad_norm": 1.1623462595850367,
+      "learning_rate": 2e-05,
+      "loss": 0.6755,
+      "step": 101
+    },
+    {
+      "epoch": 3.232,
+      "eval_loss": 0.6791322827339172,
+      "eval_runtime": 96.5222,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 101
+    },
+    {
+      "epoch": 3.2640000000000002,
+      "grad_norm": 1.16303930181089,
+      "learning_rate": 2e-05,
+      "loss": 0.6752,
+      "step": 102
+    },
+    {
+      "epoch": 3.2640000000000002,
+      "eval_loss": 0.680637538433075,
+      "eval_runtime": 96.8789,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 102
+    },
+    {
+      "epoch": 3.296,
+      "grad_norm": 1.166028046661615,
+      "learning_rate": 2e-05,
+      "loss": 0.6732,
+      "step": 103
+    },
+    {
+      "epoch": 3.296,
+      "eval_loss": 0.6818951964378357,
+      "eval_runtime": 96.7579,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 103
+    },
+    {
+      "epoch": 3.328,
+      "grad_norm": 1.2872370423601793,
+      "learning_rate": 2e-05,
+      "loss": 0.7713,
+      "step": 104
+    },
+    {
+      "epoch": 3.328,
+      "eval_loss": 0.682328462600708,
+      "eval_runtime": 98.3142,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 104
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 1.1363822202896854,
+      "learning_rate": 2e-05,
+      "loss": 0.7429,
+      "step": 105
+    },
+    {
+      "epoch": 3.36,
+      "eval_loss": 0.6817943453788757,
+      "eval_runtime": 100.7507,
+      "eval_samples_per_second": 1.985,
+      "eval_steps_per_second": 0.496,
+      "step": 105
+    },
+    {
+      "epoch": 3.416,
+      "grad_norm": 1.2711044658075554,
+      "learning_rate": 2e-05,
+      "loss": 0.7057,
+      "step": 106
+    },
+    {
+      "epoch": 3.416,
+      "eval_loss": 0.6794378161430359,
+      "eval_runtime": 100.4096,
+      "eval_samples_per_second": 1.992,
+      "eval_steps_per_second": 0.498,
+      "step": 106
+    },
+    {
+      "epoch": 3.448,
+      "grad_norm": 1.3212295597772596,
+      "learning_rate": 2e-05,
+      "loss": 0.6982,
+      "step": 107
+    },
+    {
+      "epoch": 3.448,
+      "eval_loss": 0.6767404079437256,
+      "eval_runtime": 96.6386,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 107
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 1.2420948881737728,
+      "learning_rate": 2e-05,
+      "loss": 0.7092,
+      "step": 108
+    },
+    {
+      "epoch": 3.48,
+      "eval_loss": 0.6751566529273987,
+      "eval_runtime": 95.1851,
+      "eval_samples_per_second": 2.101,
+      "eval_steps_per_second": 0.525,
+      "step": 108
+    },
+    {
+      "epoch": 3.512,
+      "grad_norm": 1.2965352636029341,
+      "learning_rate": 2e-05,
+      "loss": 0.6715,
+      "step": 109
+    },
+    {
+      "epoch": 3.512,
+      "eval_loss": 0.6750080585479736,
+      "eval_runtime": 95.3479,
+      "eval_samples_per_second": 2.098,
+      "eval_steps_per_second": 0.524,
+      "step": 109
+    },
+    {
+      "epoch": 3.544,
+      "grad_norm": 1.2789534479099607,
+      "learning_rate": 2e-05,
+      "loss": 0.6732,
+      "step": 110
+    },
+    {
+      "epoch": 3.544,
+      "eval_loss": 0.6744334697723389,
+      "eval_runtime": 95.3541,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 110
+    },
+    {
+      "epoch": 3.576,
+      "grad_norm": 1.3881035995379567,
+      "learning_rate": 2e-05,
+      "loss": 0.6777,
+      "step": 111
+    },
+    {
+      "epoch": 3.576,
+      "eval_loss": 0.6730498671531677,
+      "eval_runtime": 96.972,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.516,
+      "step": 111
+    },
+    {
+      "epoch": 3.608,
+      "grad_norm": 1.2398078245019133,
+      "learning_rate": 2e-05,
+      "loss": 0.6314,
+      "step": 112
+    },
+    {
+      "epoch": 3.608,
+      "eval_loss": 0.6725335717201233,
+      "eval_runtime": 97.1794,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.515,
+      "step": 112
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.3383993031041075,
+      "learning_rate": 2e-05,
+      "loss": 0.6776,
+      "step": 113
+    },
+    {
+      "epoch": 3.64,
+      "eval_loss": 0.6719880104064941,
+      "eval_runtime": 97.2319,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 113
+    },
+    {
+      "epoch": 3.672,
+      "grad_norm": 1.289557205987285,
+      "learning_rate": 2e-05,
+      "loss": 0.7475,
+      "step": 114
+    },
+    {
+      "epoch": 3.672,
+      "eval_loss": 0.6723533868789673,
+      "eval_runtime": 97.3108,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 114
+    },
+    {
+      "epoch": 3.7039999999999997,
+      "grad_norm": 1.121644844950096,
+      "learning_rate": 2e-05,
+      "loss": 0.6813,
+      "step": 115
+    },
+    {
+      "epoch": 3.7039999999999997,
+      "eval_loss": 0.673932671546936,
+      "eval_runtime": 97.0919,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 115
+    },
+    {
+      "epoch": 3.7359999999999998,
+      "grad_norm": 1.5370551636574723,
+      "learning_rate": 2e-05,
+      "loss": 0.7036,
+      "step": 116
+    },
+    {
+      "epoch": 3.7359999999999998,
+      "eval_loss": 0.6737978458404541,
+      "eval_runtime": 97.0064,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.515,
+      "step": 116
+    },
+    {
+      "epoch": 3.768,
+      "grad_norm": 1.3017720066449985,
+      "learning_rate": 2e-05,
+      "loss": 0.7048,
+      "step": 117
+    },
+    {
+      "epoch": 3.768,
+      "eval_loss": 0.6731936931610107,
+      "eval_runtime": 96.9134,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 117
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.3976157986974596,
+      "learning_rate": 2e-05,
+      "loss": 0.6512,
+      "step": 118
+    },
+    {
+      "epoch": 3.8,
+      "eval_loss": 0.6716210842132568,
+      "eval_runtime": 97.4028,
+      "eval_samples_per_second": 2.053,
+      "eval_steps_per_second": 0.513,
+      "step": 118
+    },
+    {
+      "epoch": 3.832,
+      "grad_norm": 1.3982632064891347,
+      "learning_rate": 2e-05,
+      "loss": 0.6481,
+      "step": 119
+    },
+    {
+      "epoch": 3.832,
+      "eval_loss": 0.6690347194671631,
+      "eval_runtime": 97.2096,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 119
+    },
+    {
+      "epoch": 3.864,
+      "grad_norm": 1.346077433712507,
+      "learning_rate": 2e-05,
+      "loss": 0.6005,
+      "step": 120
+    },
+    {
+      "epoch": 3.864,
+      "eval_loss": 0.667186975479126,
+      "eval_runtime": 97.326,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 120
+    },
+    {
+      "epoch": 3.896,
+      "grad_norm": 1.3142652448176726,
+      "learning_rate": 2e-05,
+      "loss": 0.6967,
+      "step": 121
+    },
+    {
+      "epoch": 3.896,
+      "eval_loss": 0.6662881970405579,
+      "eval_runtime": 101.7648,
+      "eval_samples_per_second": 1.965,
+      "eval_steps_per_second": 0.491,
+      "step": 121
+    },
+    {
+      "epoch": 3.928,
+      "grad_norm": 1.3368669812290557,
+      "learning_rate": 2e-05,
+      "loss": 0.6954,
+      "step": 122
+    },
+    {
+      "epoch": 3.928,
+      "eval_loss": 0.6659301519393921,
+      "eval_runtime": 98.3123,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 122
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.3992228513792793,
+      "learning_rate": 2e-05,
+      "loss": 0.6168,
+      "step": 123
+    },
+    {
+      "epoch": 3.96,
+      "eval_loss": 0.6662861704826355,
+      "eval_runtime": 98.5527,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 123
+    },
+    {
+      "epoch": 3.992,
+      "grad_norm": 1.403303492690172,
+      "learning_rate": 2e-05,
+      "loss": 0.6818,
+      "step": 124
+    },
+    {
+      "epoch": 3.992,
+      "eval_loss": 0.666582465171814,
+      "eval_runtime": 95.7576,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 124
+    },
+    {
+      "epoch": 4.024,
+      "grad_norm": 1.37383602466541,
+      "learning_rate": 2e-05,
+      "loss": 0.6539,
+      "step": 125
+    },
+    {
+      "epoch": 4.024,
+      "eval_loss": 0.6666322350502014,
+      "eval_runtime": 95.5774,
+      "eval_samples_per_second": 2.093,
+      "eval_steps_per_second": 0.523,
+      "step": 125
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 186,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 6,
+  "save_steps": 5,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": false
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 109139403341824.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-125/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c699d5bf141273a7d15c9e276f19e65cb368f8a4c2d8c5f4f59f908f45da37f8
+size 8184

checkpoint-125/zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

checkpoint-248/README.md

@@ -0,0 +1,202 @@
+---
+base_model: liuhaotian/llava-v1.6-mistral-7b
+library_name: peft
+---
+
+# 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 0.13.2

checkpoint-248/adapter_config.json

@@ -0,0 +1,34 @@
+{
+  "alpha_pattern": {},
+  "auto_mapping": null,
+  "base_model_name_or_path": "liuhaotian/llava-v1.6-mistral-7b",
+  "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": 16,
+  "lora_dropout": 0.05,
+  "megatron_config": null,
+  "megatron_core": "megatron.core",
+  "modules_to_save": null,
+  "peft_type": "LORA",
+  "r": 8,
+  "rank_pattern": {},
+  "revision": null,
+  "target_modules": [
+    "k_proj",
+    "up_proj",
+    "q_proj",
+    "down_proj",
+    "gate_proj",
+    "v_proj",
+    "o_proj"
+  ],
+  "task_type": "CAUSAL_LM",
+  "use_dora": false,
+  "use_rslora": false
+}

checkpoint-248/adapter_model.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:49c97dab65ef183af70a1f608e9dc84d43b597eb9c1c492ebcef92ad90f8c97f
+size 44387440

checkpoint-248/global_step248/zero_pp_rank_0_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5112d61ac28f472ff900463544d792bf539d41a7a8b812de1c9fc35e058742ba
+size 663858

checkpoint-248/global_step248/zero_pp_rank_0_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:509f1d7cae17905858e791ef842ad34a1c5e6964d2aaf9864979bb0194a2b63b
+size 258789357

checkpoint-248/global_step248/zero_pp_rank_1_mp_rank_00_model_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:84310f26d2a29fe71b015662bad637a324c684bdb8b2df20afbc6b97908b23ee
+size 663858

checkpoint-248/global_step248/zero_pp_rank_1_mp_rank_00_optim_states.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:73e650d7aed3c21ca69b5bf51af900f3e992bbe3c33a8f508c1c57c4b062dd78
+size 258789357

checkpoint-248/latest

@@ -0,0 +1 @@
+global_step248

checkpoint-248/rng_state_0.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f0523245ab98a0d9c42984e954d65bcd1896eb38a3c35ee40aa19fbabbdd5f6
+size 14512

checkpoint-248/rng_state_1.pth

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:895ad4a53a5764a87f970c5c06b4939bf66b3929c542fd49692094eff34e4866
+size 14512

checkpoint-248/special_tokens_map.json

@@ -0,0 +1,24 @@
+{
+  "bos_token": {
+    "content": "<s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "eos_token": {
+    "content": "</s>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": "<unk>",
+  "unk_token": {
+    "content": "<unk>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

checkpoint-248/tokenizer.model

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dadfd56d766715c61d2ef780a525ab43b8e6da4de6865bda3d95fdef5e134055
+size 493443

checkpoint-248/tokenizer_config.json

@@ -0,0 +1,45 @@
+{
+  "add_bos_token": true,
+  "add_eos_token": false,
+  "add_prefix_space": true,
+  "added_tokens_decoder": {
+    "0": {
+      "content": "<unk>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "1": {
+      "content": "<s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "2": {
+      "content": "</s>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "additional_special_tokens": [],
+  "bos_token": "<s>",
+  "chat_template": "{{ bos_token }}{% for message in messages %}{% if (message['role'] == 'user') != (loop.index0 % 2 == 0) %}{{ raise_exception('Conversation roles must alternate user/assistant/user/assistant/...') }}{% endif %}{% if message['role'] == 'user' %}{{ '[INST] ' + message['content'] + ' [/INST]' }}{% elif message['role'] == 'assistant' %}{{ message['content'] + eos_token}}{% else %}{{ raise_exception('Only user and assistant roles are supported!') }}{% endif %}{% endfor %}",
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "</s>",
+  "legacy": true,
+  "model_max_length": 4096,
+  "pad_token": "<unk>",
+  "padding_side": "right",
+  "sp_model_kwargs": {},
+  "spaces_between_special_tokens": false,
+  "tokenizer_class": "LlamaTokenizer",
+  "unk_token": "<unk>",
+  "use_default_system_prompt": false
+}

checkpoint-248/trainer_state.json

@@ -0,0 +1,3753 @@
+{
+  "best_metric": 0.6666322350502014,
+  "best_model_checkpoint": "./checkpoints/llava-v1.6-mistral/checkpoint-125",
+  "epoch": 7.992,
+  "eval_steps": 1.0,
+  "global_step": 248,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.032,
+      "grad_norm": 1.3562940564009667,
+      "learning_rate": 0.0,
+      "loss": 1.2069,
+      "step": 1
+    },
+    {
+      "epoch": 0.032,
+      "eval_loss": 1.189530849456787,
+      "eval_runtime": 100.6384,
+      "eval_samples_per_second": 1.987,
+      "eval_steps_per_second": 0.497,
+      "step": 1
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 1.3726335882966116,
+      "learning_rate": 7.737056144690831e-06,
+      "loss": 1.242,
+      "step": 2
+    },
+    {
+      "epoch": 0.064,
+      "eval_loss": 1.189530849456787,
+      "eval_runtime": 95.3713,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 2
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 1.267071497789874,
+      "learning_rate": 1.2262943855309169e-05,
+      "loss": 1.1588,
+      "step": 3
+    },
+    {
+      "epoch": 0.096,
+      "eval_loss": 1.1675384044647217,
+      "eval_runtime": 99.4768,
+      "eval_samples_per_second": 2.011,
+      "eval_steps_per_second": 0.503,
+      "step": 3
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 1.2066781722006759,
+      "learning_rate": 1.5474112289381662e-05,
+      "loss": 1.1196,
+      "step": 4
+    },
+    {
+      "epoch": 0.128,
+      "eval_loss": 1.1391011476516724,
+      "eval_runtime": 95.4777,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 4
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2746210039866817,
+      "learning_rate": 1.7964888034078543e-05,
+      "loss": 1.1462,
+      "step": 5
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 1.1023523807525635,
+      "eval_runtime": 95.4138,
+      "eval_samples_per_second": 2.096,
+      "eval_steps_per_second": 0.524,
+      "step": 5
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 1.0552650848120237,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.1147,
+      "step": 6
+    },
+    {
+      "epoch": 0.192,
+      "eval_loss": 1.063745379447937,
+      "eval_runtime": 97.6633,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 6
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 1.0779287740182022,
+      "learning_rate": 2e-05,
+      "loss": 1.1773,
+      "step": 7
+    },
+    {
+      "epoch": 0.224,
+      "eval_loss": 1.024451732635498,
+      "eval_runtime": 97.388,
+      "eval_samples_per_second": 2.054,
+      "eval_steps_per_second": 0.513,
+      "step": 7
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 0.9918200495785524,
+      "learning_rate": 2e-05,
+      "loss": 1.0434,
+      "step": 8
+    },
+    {
+      "epoch": 0.256,
+      "eval_loss": 0.9906812906265259,
+      "eval_runtime": 97.2423,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 8
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 0.9119239308208719,
+      "learning_rate": 2e-05,
+      "loss": 1.0527,
+      "step": 9
+    },
+    {
+      "epoch": 0.288,
+      "eval_loss": 0.9626247882843018,
+      "eval_runtime": 97.2048,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.514,
+      "step": 9
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.7292073828818868,
+      "learning_rate": 2e-05,
+      "loss": 0.9801,
+      "step": 10
+    },
+    {
+      "epoch": 0.32,
+      "eval_loss": 0.9422003030776978,
+      "eval_runtime": 97.143,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 10
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 0.8195137858461904,
+      "learning_rate": 2e-05,
+      "loss": 1.0104,
+      "step": 11
+    },
+    {
+      "epoch": 0.352,
+      "eval_loss": 0.9264165759086609,
+      "eval_runtime": 97.2279,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 11
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 0.7060464565699565,
+      "learning_rate": 2e-05,
+      "loss": 0.9764,
+      "step": 12
+    },
+    {
+      "epoch": 0.384,
+      "eval_loss": 0.9138455986976624,
+      "eval_runtime": 97.0954,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 12
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 0.8207130544099034,
+      "learning_rate": 2e-05,
+      "loss": 1.0371,
+      "step": 13
+    },
+    {
+      "epoch": 0.416,
+      "eval_loss": 0.902658224105835,
+      "eval_runtime": 100.8924,
+      "eval_samples_per_second": 1.982,
+      "eval_steps_per_second": 0.496,
+      "step": 13
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 0.7962544586919155,
+      "learning_rate": 2e-05,
+      "loss": 0.9455,
+      "step": 14
+    },
+    {
+      "epoch": 0.448,
+      "eval_loss": 0.8926536440849304,
+      "eval_runtime": 100.856,
+      "eval_samples_per_second": 1.983,
+      "eval_steps_per_second": 0.496,
+      "step": 14
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.8246421268957655,
+      "learning_rate": 2e-05,
+      "loss": 0.9456,
+      "step": 15
+    },
+    {
+      "epoch": 0.48,
+      "eval_loss": 0.8823295831680298,
+      "eval_runtime": 97.0568,
+      "eval_samples_per_second": 2.061,
+      "eval_steps_per_second": 0.515,
+      "step": 15
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.7040845257818263,
+      "learning_rate": 2e-05,
+      "loss": 0.9283,
+      "step": 16
+    },
+    {
+      "epoch": 0.512,
+      "eval_loss": 0.8721897602081299,
+      "eval_runtime": 98.318,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 16
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 0.7904707927869743,
+      "learning_rate": 2e-05,
+      "loss": 0.9786,
+      "step": 17
+    },
+    {
+      "epoch": 0.544,
+      "eval_loss": 0.861838698387146,
+      "eval_runtime": 101.4487,
+      "eval_samples_per_second": 1.971,
+      "eval_steps_per_second": 0.493,
+      "step": 17
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 0.7605583546638608,
+      "learning_rate": 2e-05,
+      "loss": 0.9361,
+      "step": 18
+    },
+    {
+      "epoch": 0.576,
+      "eval_loss": 0.8519415259361267,
+      "eval_runtime": 98.2448,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 18
+    },
+    {
+      "epoch": 0.608,
+      "grad_norm": 0.7005232039249588,
+      "learning_rate": 2e-05,
+      "loss": 0.8707,
+      "step": 19
+    },
+    {
+      "epoch": 0.608,
+      "eval_loss": 0.842995822429657,
+      "eval_runtime": 98.3165,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 19
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.653693259495668,
+      "learning_rate": 2e-05,
+      "loss": 0.9332,
+      "step": 20
+    },
+    {
+      "epoch": 0.64,
+      "eval_loss": 0.8355565071105957,
+      "eval_runtime": 98.237,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 20
+    },
+    {
+      "epoch": 0.672,
+      "grad_norm": 0.6600451404133434,
+      "learning_rate": 2e-05,
+      "loss": 0.8848,
+      "step": 21
+    },
+    {
+      "epoch": 0.672,
+      "eval_loss": 0.8297473788261414,
+      "eval_runtime": 98.1825,
+      "eval_samples_per_second": 2.037,
+      "eval_steps_per_second": 0.509,
+      "step": 21
+    },
+    {
+      "epoch": 0.704,
+      "grad_norm": 0.6666525650519819,
+      "learning_rate": 2e-05,
+      "loss": 0.9261,
+      "step": 22
+    },
+    {
+      "epoch": 0.704,
+      "eval_loss": 0.8248497247695923,
+      "eval_runtime": 98.1541,
+      "eval_samples_per_second": 2.038,
+      "eval_steps_per_second": 0.509,
+      "step": 22
+    },
+    {
+      "epoch": 0.736,
+      "grad_norm": 0.6759718697778233,
+      "learning_rate": 2e-05,
+      "loss": 1.0103,
+      "step": 23
+    },
+    {
+      "epoch": 0.736,
+      "eval_loss": 0.8214980959892273,
+      "eval_runtime": 98.2494,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 23
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.651870418762904,
+      "learning_rate": 2e-05,
+      "loss": 0.9538,
+      "step": 24
+    },
+    {
+      "epoch": 0.768,
+      "eval_loss": 0.818403959274292,
+      "eval_runtime": 97.9476,
+      "eval_samples_per_second": 2.042,
+      "eval_steps_per_second": 0.51,
+      "step": 24
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.6985872283132858,
+      "learning_rate": 2e-05,
+      "loss": 0.8216,
+      "step": 25
+    },
+    {
+      "epoch": 0.8,
+      "eval_loss": 0.8149588108062744,
+      "eval_runtime": 95.6175,
+      "eval_samples_per_second": 2.092,
+      "eval_steps_per_second": 0.523,
+      "step": 25
+    },
+    {
+      "epoch": 0.832,
+      "grad_norm": 0.7801182369335715,
+      "learning_rate": 2e-05,
+      "loss": 0.8478,
+      "step": 26
+    },
+    {
+      "epoch": 0.832,
+      "eval_loss": 0.8100517392158508,
+      "eval_runtime": 99.7733,
+      "eval_samples_per_second": 2.005,
+      "eval_steps_per_second": 0.501,
+      "step": 26
+    },
+    {
+      "epoch": 0.864,
+      "grad_norm": 0.6727766374526198,
+      "learning_rate": 2e-05,
+      "loss": 0.9015,
+      "step": 27
+    },
+    {
+      "epoch": 0.864,
+      "eval_loss": 0.8051578998565674,
+      "eval_runtime": 95.7387,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 27
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.7398381040653764,
+      "learning_rate": 2e-05,
+      "loss": 0.8434,
+      "step": 28
+    },
+    {
+      "epoch": 0.896,
+      "eval_loss": 0.7998965382575989,
+      "eval_runtime": 95.7753,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 28
+    },
+    {
+      "epoch": 0.928,
+      "grad_norm": 0.6837341164482282,
+      "learning_rate": 2e-05,
+      "loss": 0.9124,
+      "step": 29
+    },
+    {
+      "epoch": 0.928,
+      "eval_loss": 0.7946869134902954,
+      "eval_runtime": 95.933,
+      "eval_samples_per_second": 2.085,
+      "eval_steps_per_second": 0.521,
+      "step": 29
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.7285508052728015,
+      "learning_rate": 2e-05,
+      "loss": 0.8252,
+      "step": 30
+    },
+    {
+      "epoch": 0.96,
+      "eval_loss": 0.7891057729721069,
+      "eval_runtime": 95.871,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.522,
+      "step": 30
+    },
+    {
+      "epoch": 0.992,
+      "grad_norm": 0.706524788728722,
+      "learning_rate": 2e-05,
+      "loss": 0.8382,
+      "step": 31
+    },
+    {
+      "epoch": 0.992,
+      "eval_loss": 0.7835636138916016,
+      "eval_runtime": 95.7491,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 31
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.7392763969473979,
+      "learning_rate": 2e-05,
+      "loss": 0.8586,
+      "step": 32
+    },
+    {
+      "epoch": 1.024,
+      "eval_loss": 0.7780101299285889,
+      "eval_runtime": 99.7493,
+      "eval_samples_per_second": 2.005,
+      "eval_steps_per_second": 0.501,
+      "step": 32
+    },
+    {
+      "epoch": 1.056,
+      "grad_norm": 0.7407143277408553,
+      "learning_rate": 2e-05,
+      "loss": 0.8438,
+      "step": 33
+    },
+    {
+      "epoch": 1.056,
+      "eval_loss": 0.7728690505027771,
+      "eval_runtime": 95.8863,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.521,
+      "step": 33
+    },
+    {
+      "epoch": 1.088,
+      "grad_norm": 0.7146296362894146,
+      "learning_rate": 2e-05,
+      "loss": 0.8357,
+      "step": 34
+    },
+    {
+      "epoch": 1.088,
+      "eval_loss": 0.7683935761451721,
+      "eval_runtime": 97.824,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 34
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.7117142678212836,
+      "learning_rate": 2e-05,
+      "loss": 0.7884,
+      "step": 35
+    },
+    {
+      "epoch": 1.12,
+      "eval_loss": 0.7646369338035583,
+      "eval_runtime": 95.8185,
+      "eval_samples_per_second": 2.087,
+      "eval_steps_per_second": 0.522,
+      "step": 35
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.7735767817076163,
+      "learning_rate": 2e-05,
+      "loss": 0.8851,
+      "step": 36
+    },
+    {
+      "epoch": 1.152,
+      "eval_loss": 0.7616337537765503,
+      "eval_runtime": 96.1525,
+      "eval_samples_per_second": 2.08,
+      "eval_steps_per_second": 0.52,
+      "step": 36
+    },
+    {
+      "epoch": 1.184,
+      "grad_norm": 0.7386952203362822,
+      "learning_rate": 2e-05,
+      "loss": 0.7943,
+      "step": 37
+    },
+    {
+      "epoch": 1.184,
+      "eval_loss": 0.759408712387085,
+      "eval_runtime": 96.2772,
+      "eval_samples_per_second": 2.077,
+      "eval_steps_per_second": 0.519,
+      "step": 37
+    },
+    {
+      "epoch": 1.216,
+      "grad_norm": 0.6688305832985106,
+      "learning_rate": 2e-05,
+      "loss": 0.8433,
+      "step": 38
+    },
+    {
+      "epoch": 1.216,
+      "eval_loss": 0.7577520608901978,
+      "eval_runtime": 95.7726,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 38
+    },
+    {
+      "epoch": 1.248,
+      "grad_norm": 0.6861808214947546,
+      "learning_rate": 2e-05,
+      "loss": 0.8132,
+      "step": 39
+    },
+    {
+      "epoch": 1.248,
+      "eval_loss": 0.7565059065818787,
+      "eval_runtime": 96.1577,
+      "eval_samples_per_second": 2.08,
+      "eval_steps_per_second": 0.52,
+      "step": 39
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.7450886292190618,
+      "learning_rate": 2e-05,
+      "loss": 0.9067,
+      "step": 40
+    },
+    {
+      "epoch": 1.28,
+      "eval_loss": 0.7562046647071838,
+      "eval_runtime": 95.9983,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 40
+    },
+    {
+      "epoch": 1.312,
+      "grad_norm": 0.7508586498301053,
+      "learning_rate": 2e-05,
+      "loss": 0.8642,
+      "step": 41
+    },
+    {
+      "epoch": 1.312,
+      "eval_loss": 0.755351722240448,
+      "eval_runtime": 96.6244,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 41
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "grad_norm": 0.7717273184776977,
+      "learning_rate": 2e-05,
+      "loss": 0.9406,
+      "step": 42
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "eval_loss": 0.753773033618927,
+      "eval_runtime": 100.8127,
+      "eval_samples_per_second": 1.984,
+      "eval_steps_per_second": 0.496,
+      "step": 42
+    },
+    {
+      "epoch": 1.376,
+      "grad_norm": 0.7980115302909253,
+      "learning_rate": 2e-05,
+      "loss": 0.7732,
+      "step": 43
+    },
+    {
+      "epoch": 1.376,
+      "eval_loss": 0.7511720061302185,
+      "eval_runtime": 97.4372,
+      "eval_samples_per_second": 2.053,
+      "eval_steps_per_second": 0.513,
+      "step": 43
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.794554904521861,
+      "learning_rate": 2e-05,
+      "loss": 0.8962,
+      "step": 44
+    },
+    {
+      "epoch": 1.408,
+      "eval_loss": 0.7484354376792908,
+      "eval_runtime": 96.7621,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 44
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.7964935137573818,
+      "learning_rate": 2e-05,
+      "loss": 0.7725,
+      "step": 45
+    },
+    {
+      "epoch": 1.44,
+      "eval_loss": 0.7454251050949097,
+      "eval_runtime": 96.5292,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 45
+    },
+    {
+      "epoch": 1.472,
+      "grad_norm": 0.7477437094752549,
+      "learning_rate": 2e-05,
+      "loss": 0.8238,
+      "step": 46
+    },
+    {
+      "epoch": 1.472,
+      "eval_loss": 0.7427694201469421,
+      "eval_runtime": 96.4788,
+      "eval_samples_per_second": 2.073,
+      "eval_steps_per_second": 0.518,
+      "step": 46
+    },
+    {
+      "epoch": 1.504,
+      "grad_norm": 0.7443339980176984,
+      "learning_rate": 2e-05,
+      "loss": 0.8273,
+      "step": 47
+    },
+    {
+      "epoch": 1.504,
+      "eval_loss": 0.7407447099685669,
+      "eval_runtime": 96.5686,
+      "eval_samples_per_second": 2.071,
+      "eval_steps_per_second": 0.518,
+      "step": 47
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.7901404920428264,
+      "learning_rate": 2e-05,
+      "loss": 0.7974,
+      "step": 48
+    },
+    {
+      "epoch": 1.536,
+      "eval_loss": 0.7381842136383057,
+      "eval_runtime": 96.6071,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.518,
+      "step": 48
+    },
+    {
+      "epoch": 1.568,
+      "grad_norm": 0.8677109521747711,
+      "learning_rate": 2e-05,
+      "loss": 0.8011,
+      "step": 49
+    },
+    {
+      "epoch": 1.568,
+      "eval_loss": 0.73555588722229,
+      "eval_runtime": 96.4189,
+      "eval_samples_per_second": 2.074,
+      "eval_steps_per_second": 0.519,
+      "step": 49
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.8170436579363425,
+      "learning_rate": 2e-05,
+      "loss": 0.7762,
+      "step": 50
+    },
+    {
+      "epoch": 1.6,
+      "eval_loss": 0.7331439256668091,
+      "eval_runtime": 100.5875,
+      "eval_samples_per_second": 1.988,
+      "eval_steps_per_second": 0.497,
+      "step": 50
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "grad_norm": 0.8190539924095326,
+      "learning_rate": 2e-05,
+      "loss": 0.7469,
+      "step": 51
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "eval_loss": 0.7306647300720215,
+      "eval_runtime": 101.1429,
+      "eval_samples_per_second": 1.977,
+      "eval_steps_per_second": 0.494,
+      "step": 51
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.8044932603192197,
+      "learning_rate": 2e-05,
+      "loss": 0.8052,
+      "step": 52
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "eval_loss": 0.7287429571151733,
+      "eval_runtime": 97.25,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 52
+    },
+    {
+      "epoch": 1.696,
+      "grad_norm": 0.7811170846407103,
+      "learning_rate": 2e-05,
+      "loss": 0.755,
+      "step": 53
+    },
+    {
+      "epoch": 1.696,
+      "eval_loss": 0.7264651656150818,
+      "eval_runtime": 97.67,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 53
+    },
+    {
+      "epoch": 1.728,
+      "grad_norm": 0.808405093976468,
+      "learning_rate": 2e-05,
+      "loss": 0.7838,
+      "step": 54
+    },
+    {
+      "epoch": 1.728,
+      "eval_loss": 0.7242828607559204,
+      "eval_runtime": 101.2008,
+      "eval_samples_per_second": 1.976,
+      "eval_steps_per_second": 0.494,
+      "step": 54
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.8172106270954758,
+      "learning_rate": 2e-05,
+      "loss": 0.7123,
+      "step": 55
+    },
+    {
+      "epoch": 1.76,
+      "eval_loss": 0.7217574119567871,
+      "eval_runtime": 99.6145,
+      "eval_samples_per_second": 2.008,
+      "eval_steps_per_second": 0.502,
+      "step": 55
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.83212530473105,
+      "learning_rate": 2e-05,
+      "loss": 0.8141,
+      "step": 56
+    },
+    {
+      "epoch": 1.792,
+      "eval_loss": 0.7192932367324829,
+      "eval_runtime": 97.1377,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 56
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "grad_norm": 0.8632612959683078,
+      "learning_rate": 2e-05,
+      "loss": 0.8005,
+      "step": 57
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "eval_loss": 0.717003345489502,
+      "eval_runtime": 97.2871,
+      "eval_samples_per_second": 2.056,
+      "eval_steps_per_second": 0.514,
+      "step": 57
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "grad_norm": 0.8362081570841255,
+      "learning_rate": 2e-05,
+      "loss": 0.8254,
+      "step": 58
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "eval_loss": 0.715446949005127,
+      "eval_runtime": 97.5751,
+      "eval_samples_per_second": 2.05,
+      "eval_steps_per_second": 0.512,
+      "step": 58
+    },
+    {
+      "epoch": 1.888,
+      "grad_norm": 0.8862779841020042,
+      "learning_rate": 2e-05,
+      "loss": 0.7949,
+      "step": 59
+    },
+    {
+      "epoch": 1.888,
+      "eval_loss": 0.7145370244979858,
+      "eval_runtime": 97.2406,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 59
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.875656256488838,
+      "learning_rate": 2e-05,
+      "loss": 0.8277,
+      "step": 60
+    },
+    {
+      "epoch": 1.92,
+      "eval_loss": 0.7140547037124634,
+      "eval_runtime": 97.1485,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 60
+    },
+    {
+      "epoch": 1.952,
+      "grad_norm": 0.8850690645543244,
+      "learning_rate": 2e-05,
+      "loss": 0.8117,
+      "step": 61
+    },
+    {
+      "epoch": 1.952,
+      "eval_loss": 0.7128701210021973,
+      "eval_runtime": 97.8906,
+      "eval_samples_per_second": 2.043,
+      "eval_steps_per_second": 0.511,
+      "step": 61
+    },
+    {
+      "epoch": 1.984,
+      "grad_norm": 0.8961641824408927,
+      "learning_rate": 2e-05,
+      "loss": 0.704,
+      "step": 62
+    },
+    {
+      "epoch": 1.984,
+      "eval_loss": 0.7107064127922058,
+      "eval_runtime": 102.2163,
+      "eval_samples_per_second": 1.957,
+      "eval_steps_per_second": 0.489,
+      "step": 62
+    },
+    {
+      "epoch": 2.016,
+      "grad_norm": 0.8602247372004996,
+      "learning_rate": 2e-05,
+      "loss": 0.8016,
+      "step": 63
+    },
+    {
+      "epoch": 2.016,
+      "eval_loss": 0.7091581225395203,
+      "eval_runtime": 98.2114,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 63
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.7881703384753784,
+      "learning_rate": 2e-05,
+      "loss": 0.7912,
+      "step": 64
+    },
+    {
+      "epoch": 2.048,
+      "eval_loss": 0.7081906199455261,
+      "eval_runtime": 98.9403,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 64
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.8436680506716614,
+      "learning_rate": 2e-05,
+      "loss": 0.6965,
+      "step": 65
+    },
+    {
+      "epoch": 2.08,
+      "eval_loss": 0.7070262432098389,
+      "eval_runtime": 97.7451,
+      "eval_samples_per_second": 2.046,
+      "eval_steps_per_second": 0.512,
+      "step": 65
+    },
+    {
+      "epoch": 2.112,
+      "grad_norm": 0.8694446846234115,
+      "learning_rate": 2e-05,
+      "loss": 0.7015,
+      "step": 66
+    },
+    {
+      "epoch": 2.112,
+      "eval_loss": 0.7055197358131409,
+      "eval_runtime": 97.994,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 66
+    },
+    {
+      "epoch": 2.144,
+      "grad_norm": 0.9518475022730614,
+      "learning_rate": 2e-05,
+      "loss": 0.6952,
+      "step": 67
+    },
+    {
+      "epoch": 2.144,
+      "eval_loss": 0.7035844326019287,
+      "eval_runtime": 98.0296,
+      "eval_samples_per_second": 2.04,
+      "eval_steps_per_second": 0.51,
+      "step": 67
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.8662066024104106,
+      "learning_rate": 2e-05,
+      "loss": 0.7731,
+      "step": 68
+    },
+    {
+      "epoch": 2.176,
+      "eval_loss": 0.7019283771514893,
+      "eval_runtime": 100.0982,
+      "eval_samples_per_second": 1.998,
+      "eval_steps_per_second": 0.5,
+      "step": 68
+    },
+    {
+      "epoch": 2.208,
+      "grad_norm": 0.9574631942368209,
+      "learning_rate": 2e-05,
+      "loss": 0.7572,
+      "step": 69
+    },
+    {
+      "epoch": 2.208,
+      "eval_loss": 0.7011401653289795,
+      "eval_runtime": 101.9267,
+      "eval_samples_per_second": 1.962,
+      "eval_steps_per_second": 0.491,
+      "step": 69
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.9477542521472675,
+      "learning_rate": 2e-05,
+      "loss": 0.7393,
+      "step": 70
+    },
+    {
+      "epoch": 2.24,
+      "eval_loss": 0.7006180286407471,
+      "eval_runtime": 101.9672,
+      "eval_samples_per_second": 1.961,
+      "eval_steps_per_second": 0.49,
+      "step": 70
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "grad_norm": 0.9612956116350272,
+      "learning_rate": 2e-05,
+      "loss": 0.7175,
+      "step": 71
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "eval_loss": 0.7002539038658142,
+      "eval_runtime": 101.8755,
+      "eval_samples_per_second": 1.963,
+      "eval_steps_per_second": 0.491,
+      "step": 71
+    },
+    {
+      "epoch": 2.304,
+      "grad_norm": 0.9934844303955727,
+      "learning_rate": 2e-05,
+      "loss": 0.791,
+      "step": 72
+    },
+    {
+      "epoch": 2.304,
+      "eval_loss": 0.6997203230857849,
+      "eval_runtime": 97.9926,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 72
+    },
+    {
+      "epoch": 2.336,
+      "grad_norm": 1.0540430431227044,
+      "learning_rate": 2e-05,
+      "loss": 0.7542,
+      "step": 73
+    },
+    {
+      "epoch": 2.336,
+      "eval_loss": 0.6988361477851868,
+      "eval_runtime": 100.3704,
+      "eval_samples_per_second": 1.993,
+      "eval_steps_per_second": 0.498,
+      "step": 73
+    },
+    {
+      "epoch": 2.368,
+      "grad_norm": 1.0249397957961794,
+      "learning_rate": 2e-05,
+      "loss": 0.8084,
+      "step": 74
+    },
+    {
+      "epoch": 2.368,
+      "eval_loss": 0.6980065703392029,
+      "eval_runtime": 101.8958,
+      "eval_samples_per_second": 1.963,
+      "eval_steps_per_second": 0.491,
+      "step": 74
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.0445498365690145,
+      "learning_rate": 2e-05,
+      "loss": 0.7964,
+      "step": 75
+    },
+    {
+      "epoch": 2.4,
+      "eval_loss": 0.6971798539161682,
+      "eval_runtime": 98.1624,
+      "eval_samples_per_second": 2.037,
+      "eval_steps_per_second": 0.509,
+      "step": 75
+    },
+    {
+      "epoch": 2.432,
+      "grad_norm": 0.9685893079320761,
+      "learning_rate": 2e-05,
+      "loss": 0.8047,
+      "step": 76
+    },
+    {
+      "epoch": 2.432,
+      "eval_loss": 0.696861743927002,
+      "eval_runtime": 98.6404,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 76
+    },
+    {
+      "epoch": 2.464,
+      "grad_norm": 0.9753091933204456,
+      "learning_rate": 2e-05,
+      "loss": 0.684,
+      "step": 77
+    },
+    {
+      "epoch": 2.464,
+      "eval_loss": 0.6957904100418091,
+      "eval_runtime": 98.8766,
+      "eval_samples_per_second": 2.023,
+      "eval_steps_per_second": 0.506,
+      "step": 77
+    },
+    {
+      "epoch": 2.496,
+      "grad_norm": 0.9389149478503764,
+      "learning_rate": 2e-05,
+      "loss": 0.7434,
+      "step": 78
+    },
+    {
+      "epoch": 2.496,
+      "eval_loss": 0.6943306922912598,
+      "eval_runtime": 98.5751,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 78
+    },
+    {
+      "epoch": 2.528,
+      "grad_norm": 1.0675154845211299,
+      "learning_rate": 2e-05,
+      "loss": 0.7208,
+      "step": 79
+    },
+    {
+      "epoch": 2.528,
+      "eval_loss": 0.6920651197433472,
+      "eval_runtime": 99.0851,
+      "eval_samples_per_second": 2.018,
+      "eval_steps_per_second": 0.505,
+      "step": 79
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.9937936593307737,
+      "learning_rate": 2e-05,
+      "loss": 0.6948,
+      "step": 80
+    },
+    {
+      "epoch": 2.56,
+      "eval_loss": 0.6899142265319824,
+      "eval_runtime": 98.7974,
+      "eval_samples_per_second": 2.024,
+      "eval_steps_per_second": 0.506,
+      "step": 80
+    },
+    {
+      "epoch": 2.592,
+      "grad_norm": 0.9650832276698476,
+      "learning_rate": 2e-05,
+      "loss": 0.7666,
+      "step": 81
+    },
+    {
+      "epoch": 2.592,
+      "eval_loss": 0.6886695623397827,
+      "eval_runtime": 98.6813,
+      "eval_samples_per_second": 2.027,
+      "eval_steps_per_second": 0.507,
+      "step": 81
+    },
+    {
+      "epoch": 2.624,
+      "grad_norm": 0.9961610958296112,
+      "learning_rate": 2e-05,
+      "loss": 0.7741,
+      "step": 82
+    },
+    {
+      "epoch": 2.624,
+      "eval_loss": 0.687745213508606,
+      "eval_runtime": 98.624,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 82
+    },
+    {
+      "epoch": 2.656,
+      "grad_norm": 1.0261499699526089,
+      "learning_rate": 2e-05,
+      "loss": 0.7747,
+      "step": 83
+    },
+    {
+      "epoch": 2.656,
+      "eval_loss": 0.6869972944259644,
+      "eval_runtime": 98.9604,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 83
+    },
+    {
+      "epoch": 2.6879999999999997,
+      "grad_norm": 1.025059237745532,
+      "learning_rate": 2e-05,
+      "loss": 0.7526,
+      "step": 84
+    },
+    {
+      "epoch": 2.6879999999999997,
+      "eval_loss": 0.6862147450447083,
+      "eval_runtime": 98.4387,
+      "eval_samples_per_second": 2.032,
+      "eval_steps_per_second": 0.508,
+      "step": 84
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "grad_norm": 1.1383626266013125,
+      "learning_rate": 2e-05,
+      "loss": 0.683,
+      "step": 85
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_loss": 0.6845572590827942,
+      "eval_runtime": 98.4979,
+      "eval_samples_per_second": 2.03,
+      "eval_steps_per_second": 0.508,
+      "step": 85
+    },
+    {
+      "epoch": 2.752,
+      "grad_norm": 1.0427739679421295,
+      "learning_rate": 2e-05,
+      "loss": 0.7362,
+      "step": 86
+    },
+    {
+      "epoch": 2.752,
+      "eval_loss": 0.683280348777771,
+      "eval_runtime": 103.6707,
+      "eval_samples_per_second": 1.929,
+      "eval_steps_per_second": 0.482,
+      "step": 86
+    },
+    {
+      "epoch": 2.784,
+      "grad_norm": 1.1280129141879938,
+      "learning_rate": 2e-05,
+      "loss": 0.7743,
+      "step": 87
+    },
+    {
+      "epoch": 2.784,
+      "eval_loss": 0.6824235320091248,
+      "eval_runtime": 99.2995,
+      "eval_samples_per_second": 2.014,
+      "eval_steps_per_second": 0.504,
+      "step": 87
+    },
+    {
+      "epoch": 2.816,
+      "grad_norm": 1.0017715497784696,
+      "learning_rate": 2e-05,
+      "loss": 0.7164,
+      "step": 88
+    },
+    {
+      "epoch": 2.816,
+      "eval_loss": 0.6819013357162476,
+      "eval_runtime": 99.5206,
+      "eval_samples_per_second": 2.01,
+      "eval_steps_per_second": 0.502,
+      "step": 88
+    },
+    {
+      "epoch": 2.848,
+      "grad_norm": 1.0769673642284994,
+      "learning_rate": 2e-05,
+      "loss": 0.7013,
+      "step": 89
+    },
+    {
+      "epoch": 2.848,
+      "eval_loss": 0.6820746064186096,
+      "eval_runtime": 96.5197,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 89
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 1.0389167281844591,
+      "learning_rate": 2e-05,
+      "loss": 0.6805,
+      "step": 90
+    },
+    {
+      "epoch": 2.88,
+      "eval_loss": 0.6826525926589966,
+      "eval_runtime": 96.0942,
+      "eval_samples_per_second": 2.081,
+      "eval_steps_per_second": 0.52,
+      "step": 90
+    },
+    {
+      "epoch": 2.912,
+      "grad_norm": 0.9705743838620626,
+      "learning_rate": 2e-05,
+      "loss": 0.7681,
+      "step": 91
+    },
+    {
+      "epoch": 2.912,
+      "eval_loss": 0.6836435198783875,
+      "eval_runtime": 96.2441,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 91
+    },
+    {
+      "epoch": 2.944,
+      "grad_norm": 1.037823791993831,
+      "learning_rate": 2e-05,
+      "loss": 0.7532,
+      "step": 92
+    },
+    {
+      "epoch": 2.944,
+      "eval_loss": 0.6845746040344238,
+      "eval_runtime": 96.2394,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 92
+    },
+    {
+      "epoch": 2.976,
+      "grad_norm": 1.1323835942146157,
+      "learning_rate": 2e-05,
+      "loss": 0.7171,
+      "step": 93
+    },
+    {
+      "epoch": 2.976,
+      "eval_loss": 0.684663712978363,
+      "eval_runtime": 96.23,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 93
+    },
+    {
+      "epoch": 3.008,
+      "grad_norm": 1.1957864756602699,
+      "learning_rate": 2e-05,
+      "loss": 0.7166,
+      "step": 94
+    },
+    {
+      "epoch": 3.008,
+      "eval_loss": 0.6830846667289734,
+      "eval_runtime": 96.6549,
+      "eval_samples_per_second": 2.069,
+      "eval_steps_per_second": 0.517,
+      "step": 94
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 1.1077357794232636,
+      "learning_rate": 2e-05,
+      "loss": 0.6667,
+      "step": 95
+    },
+    {
+      "epoch": 3.04,
+      "eval_loss": 0.6810076832771301,
+      "eval_runtime": 96.3239,
+      "eval_samples_per_second": 2.076,
+      "eval_steps_per_second": 0.519,
+      "step": 95
+    },
+    {
+      "epoch": 3.072,
+      "grad_norm": 1.1851219157184936,
+      "learning_rate": 2e-05,
+      "loss": 0.7115,
+      "step": 96
+    },
+    {
+      "epoch": 3.072,
+      "eval_loss": 0.6796395778656006,
+      "eval_runtime": 96.9109,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 96
+    },
+    {
+      "epoch": 3.104,
+      "grad_norm": 1.0812671042616444,
+      "learning_rate": 2e-05,
+      "loss": 0.7333,
+      "step": 97
+    },
+    {
+      "epoch": 3.104,
+      "eval_loss": 0.6794567108154297,
+      "eval_runtime": 96.7403,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 97
+    },
+    {
+      "epoch": 3.136,
+      "grad_norm": 1.130095597839828,
+      "learning_rate": 2e-05,
+      "loss": 0.6328,
+      "step": 98
+    },
+    {
+      "epoch": 3.136,
+      "eval_loss": 0.6792007684707642,
+      "eval_runtime": 96.8136,
+      "eval_samples_per_second": 2.066,
+      "eval_steps_per_second": 0.516,
+      "step": 98
+    },
+    {
+      "epoch": 3.168,
+      "grad_norm": 1.16102100344116,
+      "learning_rate": 2e-05,
+      "loss": 0.6625,
+      "step": 99
+    },
+    {
+      "epoch": 3.168,
+      "eval_loss": 0.6789132952690125,
+      "eval_runtime": 96.6982,
+      "eval_samples_per_second": 2.068,
+      "eval_steps_per_second": 0.517,
+      "step": 99
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 1.226689811951201,
+      "learning_rate": 2e-05,
+      "loss": 0.7522,
+      "step": 100
+    },
+    {
+      "epoch": 3.2,
+      "eval_loss": 0.6786602735519409,
+      "eval_runtime": 96.6253,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 100
+    },
+    {
+      "epoch": 3.232,
+      "grad_norm": 1.1623462595850367,
+      "learning_rate": 2e-05,
+      "loss": 0.6755,
+      "step": 101
+    },
+    {
+      "epoch": 3.232,
+      "eval_loss": 0.6791322827339172,
+      "eval_runtime": 96.5222,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 101
+    },
+    {
+      "epoch": 3.2640000000000002,
+      "grad_norm": 1.16303930181089,
+      "learning_rate": 2e-05,
+      "loss": 0.6752,
+      "step": 102
+    },
+    {
+      "epoch": 3.2640000000000002,
+      "eval_loss": 0.680637538433075,
+      "eval_runtime": 96.8789,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 102
+    },
+    {
+      "epoch": 3.296,
+      "grad_norm": 1.166028046661615,
+      "learning_rate": 2e-05,
+      "loss": 0.6732,
+      "step": 103
+    },
+    {
+      "epoch": 3.296,
+      "eval_loss": 0.6818951964378357,
+      "eval_runtime": 96.7579,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 103
+    },
+    {
+      "epoch": 3.328,
+      "grad_norm": 1.2872370423601793,
+      "learning_rate": 2e-05,
+      "loss": 0.7713,
+      "step": 104
+    },
+    {
+      "epoch": 3.328,
+      "eval_loss": 0.682328462600708,
+      "eval_runtime": 98.3142,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 104
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 1.1363822202896854,
+      "learning_rate": 2e-05,
+      "loss": 0.7429,
+      "step": 105
+    },
+    {
+      "epoch": 3.36,
+      "eval_loss": 0.6817943453788757,
+      "eval_runtime": 100.7507,
+      "eval_samples_per_second": 1.985,
+      "eval_steps_per_second": 0.496,
+      "step": 105
+    },
+    {
+      "epoch": 3.416,
+      "grad_norm": 1.2711044658075554,
+      "learning_rate": 2e-05,
+      "loss": 0.7057,
+      "step": 106
+    },
+    {
+      "epoch": 3.416,
+      "eval_loss": 0.6794378161430359,
+      "eval_runtime": 100.4096,
+      "eval_samples_per_second": 1.992,
+      "eval_steps_per_second": 0.498,
+      "step": 106
+    },
+    {
+      "epoch": 3.448,
+      "grad_norm": 1.3212295597772596,
+      "learning_rate": 2e-05,
+      "loss": 0.6982,
+      "step": 107
+    },
+    {
+      "epoch": 3.448,
+      "eval_loss": 0.6767404079437256,
+      "eval_runtime": 96.6386,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 107
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 1.2420948881737728,
+      "learning_rate": 2e-05,
+      "loss": 0.7092,
+      "step": 108
+    },
+    {
+      "epoch": 3.48,
+      "eval_loss": 0.6751566529273987,
+      "eval_runtime": 95.1851,
+      "eval_samples_per_second": 2.101,
+      "eval_steps_per_second": 0.525,
+      "step": 108
+    },
+    {
+      "epoch": 3.512,
+      "grad_norm": 1.2965352636029341,
+      "learning_rate": 2e-05,
+      "loss": 0.6715,
+      "step": 109
+    },
+    {
+      "epoch": 3.512,
+      "eval_loss": 0.6750080585479736,
+      "eval_runtime": 95.3479,
+      "eval_samples_per_second": 2.098,
+      "eval_steps_per_second": 0.524,
+      "step": 109
+    },
+    {
+      "epoch": 3.544,
+      "grad_norm": 1.2789534479099607,
+      "learning_rate": 2e-05,
+      "loss": 0.6732,
+      "step": 110
+    },
+    {
+      "epoch": 3.544,
+      "eval_loss": 0.6744334697723389,
+      "eval_runtime": 95.3541,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 110
+    },
+    {
+      "epoch": 3.576,
+      "grad_norm": 1.3881035995379567,
+      "learning_rate": 2e-05,
+      "loss": 0.6777,
+      "step": 111
+    },
+    {
+      "epoch": 3.576,
+      "eval_loss": 0.6730498671531677,
+      "eval_runtime": 96.972,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.516,
+      "step": 111
+    },
+    {
+      "epoch": 3.608,
+      "grad_norm": 1.2398078245019133,
+      "learning_rate": 2e-05,
+      "loss": 0.6314,
+      "step": 112
+    },
+    {
+      "epoch": 3.608,
+      "eval_loss": 0.6725335717201233,
+      "eval_runtime": 97.1794,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.515,
+      "step": 112
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.3383993031041075,
+      "learning_rate": 2e-05,
+      "loss": 0.6776,
+      "step": 113
+    },
+    {
+      "epoch": 3.64,
+      "eval_loss": 0.6719880104064941,
+      "eval_runtime": 97.2319,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 113
+    },
+    {
+      "epoch": 3.672,
+      "grad_norm": 1.289557205987285,
+      "learning_rate": 2e-05,
+      "loss": 0.7475,
+      "step": 114
+    },
+    {
+      "epoch": 3.672,
+      "eval_loss": 0.6723533868789673,
+      "eval_runtime": 97.3108,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 114
+    },
+    {
+      "epoch": 3.7039999999999997,
+      "grad_norm": 1.121644844950096,
+      "learning_rate": 2e-05,
+      "loss": 0.6813,
+      "step": 115
+    },
+    {
+      "epoch": 3.7039999999999997,
+      "eval_loss": 0.673932671546936,
+      "eval_runtime": 97.0919,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 115
+    },
+    {
+      "epoch": 3.7359999999999998,
+      "grad_norm": 1.5370551636574723,
+      "learning_rate": 2e-05,
+      "loss": 0.7036,
+      "step": 116
+    },
+    {
+      "epoch": 3.7359999999999998,
+      "eval_loss": 0.6737978458404541,
+      "eval_runtime": 97.0064,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.515,
+      "step": 116
+    },
+    {
+      "epoch": 3.768,
+      "grad_norm": 1.3017720066449985,
+      "learning_rate": 2e-05,
+      "loss": 0.7048,
+      "step": 117
+    },
+    {
+      "epoch": 3.768,
+      "eval_loss": 0.6731936931610107,
+      "eval_runtime": 96.9134,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 117
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.3976157986974596,
+      "learning_rate": 2e-05,
+      "loss": 0.6512,
+      "step": 118
+    },
+    {
+      "epoch": 3.8,
+      "eval_loss": 0.6716210842132568,
+      "eval_runtime": 97.4028,
+      "eval_samples_per_second": 2.053,
+      "eval_steps_per_second": 0.513,
+      "step": 118
+    },
+    {
+      "epoch": 3.832,
+      "grad_norm": 1.3982632064891347,
+      "learning_rate": 2e-05,
+      "loss": 0.6481,
+      "step": 119
+    },
+    {
+      "epoch": 3.832,
+      "eval_loss": 0.6690347194671631,
+      "eval_runtime": 97.2096,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 119
+    },
+    {
+      "epoch": 3.864,
+      "grad_norm": 1.346077433712507,
+      "learning_rate": 2e-05,
+      "loss": 0.6005,
+      "step": 120
+    },
+    {
+      "epoch": 3.864,
+      "eval_loss": 0.667186975479126,
+      "eval_runtime": 97.326,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 120
+    },
+    {
+      "epoch": 3.896,
+      "grad_norm": 1.3142652448176726,
+      "learning_rate": 2e-05,
+      "loss": 0.6967,
+      "step": 121
+    },
+    {
+      "epoch": 3.896,
+      "eval_loss": 0.6662881970405579,
+      "eval_runtime": 101.7648,
+      "eval_samples_per_second": 1.965,
+      "eval_steps_per_second": 0.491,
+      "step": 121
+    },
+    {
+      "epoch": 3.928,
+      "grad_norm": 1.3368669812290557,
+      "learning_rate": 2e-05,
+      "loss": 0.6954,
+      "step": 122
+    },
+    {
+      "epoch": 3.928,
+      "eval_loss": 0.6659301519393921,
+      "eval_runtime": 98.3123,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 122
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.3992228513792793,
+      "learning_rate": 2e-05,
+      "loss": 0.6168,
+      "step": 123
+    },
+    {
+      "epoch": 3.96,
+      "eval_loss": 0.6662861704826355,
+      "eval_runtime": 98.5527,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 123
+    },
+    {
+      "epoch": 3.992,
+      "grad_norm": 1.403303492690172,
+      "learning_rate": 2e-05,
+      "loss": 0.6818,
+      "step": 124
+    },
+    {
+      "epoch": 3.992,
+      "eval_loss": 0.666582465171814,
+      "eval_runtime": 95.7576,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 124
+    },
+    {
+      "epoch": 4.024,
+      "grad_norm": 1.37383602466541,
+      "learning_rate": 2e-05,
+      "loss": 0.6539,
+      "step": 125
+    },
+    {
+      "epoch": 4.024,
+      "eval_loss": 0.6666322350502014,
+      "eval_runtime": 95.5774,
+      "eval_samples_per_second": 2.093,
+      "eval_steps_per_second": 0.523,
+      "step": 125
+    },
+    {
+      "epoch": 4.056,
+      "grad_norm": 1.3760569151679132,
+      "learning_rate": 2e-05,
+      "loss": 0.7157,
+      "step": 126
+    },
+    {
+      "epoch": 4.056,
+      "eval_loss": 0.6672561764717102,
+      "eval_runtime": 99.9115,
+      "eval_samples_per_second": 2.002,
+      "eval_steps_per_second": 0.5,
+      "step": 126
+    },
+    {
+      "epoch": 4.088,
+      "grad_norm": 1.3234088412698182,
+      "learning_rate": 2e-05,
+      "loss": 0.644,
+      "step": 127
+    },
+    {
+      "epoch": 4.088,
+      "eval_loss": 0.6687076091766357,
+      "eval_runtime": 95.8064,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 127
+    },
+    {
+      "epoch": 4.12,
+      "grad_norm": 1.300670493052994,
+      "learning_rate": 2e-05,
+      "loss": 0.6551,
+      "step": 128
+    },
+    {
+      "epoch": 4.12,
+      "eval_loss": 0.6709404587745667,
+      "eval_runtime": 97.4903,
+      "eval_samples_per_second": 2.051,
+      "eval_steps_per_second": 0.513,
+      "step": 128
+    },
+    {
+      "epoch": 4.152,
+      "grad_norm": 1.469380020399129,
+      "learning_rate": 2e-05,
+      "loss": 0.584,
+      "step": 129
+    },
+    {
+      "epoch": 4.152,
+      "eval_loss": 0.6740531325340271,
+      "eval_runtime": 95.7054,
+      "eval_samples_per_second": 2.09,
+      "eval_steps_per_second": 0.522,
+      "step": 129
+    },
+    {
+      "epoch": 4.184,
+      "grad_norm": 1.411712759624833,
+      "learning_rate": 2e-05,
+      "loss": 0.6349,
+      "step": 130
+    },
+    {
+      "epoch": 4.184,
+      "eval_loss": 0.676266074180603,
+      "eval_runtime": 95.9231,
+      "eval_samples_per_second": 2.085,
+      "eval_steps_per_second": 0.521,
+      "step": 130
+    },
+    {
+      "epoch": 4.216,
+      "grad_norm": 1.5130330368136387,
+      "learning_rate": 2e-05,
+      "loss": 0.6328,
+      "step": 131
+    },
+    {
+      "epoch": 4.216,
+      "eval_loss": 0.6771231293678284,
+      "eval_runtime": 95.9553,
+      "eval_samples_per_second": 2.084,
+      "eval_steps_per_second": 0.521,
+      "step": 131
+    },
+    {
+      "epoch": 4.248,
+      "grad_norm": 1.6050594949836388,
+      "learning_rate": 2e-05,
+      "loss": 0.6015,
+      "step": 132
+    },
+    {
+      "epoch": 4.248,
+      "eval_loss": 0.6763471961021423,
+      "eval_runtime": 96.0391,
+      "eval_samples_per_second": 2.082,
+      "eval_steps_per_second": 0.521,
+      "step": 132
+    },
+    {
+      "epoch": 4.28,
+      "grad_norm": 1.6386261889303306,
+      "learning_rate": 2e-05,
+      "loss": 0.6793,
+      "step": 133
+    },
+    {
+      "epoch": 4.28,
+      "eval_loss": 0.6736429333686829,
+      "eval_runtime": 96.0315,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 133
+    },
+    {
+      "epoch": 4.312,
+      "grad_norm": 1.6157081951509844,
+      "learning_rate": 2e-05,
+      "loss": 0.5718,
+      "step": 134
+    },
+    {
+      "epoch": 4.312,
+      "eval_loss": 0.672798752784729,
+      "eval_runtime": 95.7787,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 134
+    },
+    {
+      "epoch": 4.344,
+      "grad_norm": 1.5634717589762641,
+      "learning_rate": 2e-05,
+      "loss": 0.6596,
+      "step": 135
+    },
+    {
+      "epoch": 4.344,
+      "eval_loss": 0.67247474193573,
+      "eval_runtime": 96.034,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 135
+    },
+    {
+      "epoch": 4.376,
+      "grad_norm": 1.428039434695232,
+      "learning_rate": 2e-05,
+      "loss": 0.6232,
+      "step": 136
+    },
+    {
+      "epoch": 4.376,
+      "eval_loss": 0.6719217896461487,
+      "eval_runtime": 100.919,
+      "eval_samples_per_second": 1.982,
+      "eval_steps_per_second": 0.495,
+      "step": 136
+    },
+    {
+      "epoch": 4.408,
+      "grad_norm": 1.6531362233925706,
+      "learning_rate": 2e-05,
+      "loss": 0.633,
+      "step": 137
+    },
+    {
+      "epoch": 4.408,
+      "eval_loss": 0.6712325811386108,
+      "eval_runtime": 96.846,
+      "eval_samples_per_second": 2.065,
+      "eval_steps_per_second": 0.516,
+      "step": 137
+    },
+    {
+      "epoch": 4.44,
+      "grad_norm": 1.5172899677155884,
+      "learning_rate": 2e-05,
+      "loss": 0.655,
+      "step": 138
+    },
+    {
+      "epoch": 4.44,
+      "eval_loss": 0.6704159379005432,
+      "eval_runtime": 96.9878,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.516,
+      "step": 138
+    },
+    {
+      "epoch": 4.4719999999999995,
+      "grad_norm": 1.561414875887358,
+      "learning_rate": 2e-05,
+      "loss": 0.6457,
+      "step": 139
+    },
+    {
+      "epoch": 4.4719999999999995,
+      "eval_loss": 0.6703745126724243,
+      "eval_runtime": 96.8356,
+      "eval_samples_per_second": 2.065,
+      "eval_steps_per_second": 0.516,
+      "step": 139
+    },
+    {
+      "epoch": 4.504,
+      "grad_norm": 1.559805415806941,
+      "learning_rate": 2e-05,
+      "loss": 0.5771,
+      "step": 140
+    },
+    {
+      "epoch": 4.504,
+      "eval_loss": 0.6714787483215332,
+      "eval_runtime": 96.7177,
+      "eval_samples_per_second": 2.068,
+      "eval_steps_per_second": 0.517,
+      "step": 140
+    },
+    {
+      "epoch": 4.536,
+      "grad_norm": 1.4633822461746586,
+      "learning_rate": 2e-05,
+      "loss": 0.6637,
+      "step": 141
+    },
+    {
+      "epoch": 4.536,
+      "eval_loss": 0.6731461882591248,
+      "eval_runtime": 96.6195,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 141
+    },
+    {
+      "epoch": 4.568,
+      "grad_norm": 1.5001482095613168,
+      "learning_rate": 2e-05,
+      "loss": 0.599,
+      "step": 142
+    },
+    {
+      "epoch": 4.568,
+      "eval_loss": 0.6755445599555969,
+      "eval_runtime": 96.8032,
+      "eval_samples_per_second": 2.066,
+      "eval_steps_per_second": 0.517,
+      "step": 142
+    },
+    {
+      "epoch": 4.6,
+      "grad_norm": 1.5542517583357578,
+      "learning_rate": 2e-05,
+      "loss": 0.6962,
+      "step": 143
+    },
+    {
+      "epoch": 4.6,
+      "eval_loss": 0.676831841468811,
+      "eval_runtime": 96.681,
+      "eval_samples_per_second": 2.069,
+      "eval_steps_per_second": 0.517,
+      "step": 143
+    },
+    {
+      "epoch": 4.632,
+      "grad_norm": 1.7874313314626142,
+      "learning_rate": 2e-05,
+      "loss": 0.672,
+      "step": 144
+    },
+    {
+      "epoch": 4.632,
+      "eval_loss": 0.6758092641830444,
+      "eval_runtime": 96.7218,
+      "eval_samples_per_second": 2.068,
+      "eval_steps_per_second": 0.517,
+      "step": 144
+    },
+    {
+      "epoch": 4.664,
+      "grad_norm": 1.7573329016960397,
+      "learning_rate": 2e-05,
+      "loss": 0.6765,
+      "step": 145
+    },
+    {
+      "epoch": 4.664,
+      "eval_loss": 0.673170804977417,
+      "eval_runtime": 96.9733,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.516,
+      "step": 145
+    },
+    {
+      "epoch": 4.696,
+      "grad_norm": 1.810854526623386,
+      "learning_rate": 2e-05,
+      "loss": 0.603,
+      "step": 146
+    },
+    {
+      "epoch": 4.696,
+      "eval_loss": 0.6707517504692078,
+      "eval_runtime": 97.1234,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 146
+    },
+    {
+      "epoch": 4.728,
+      "grad_norm": 1.6602824802653684,
+      "learning_rate": 2e-05,
+      "loss": 0.6737,
+      "step": 147
+    },
+    {
+      "epoch": 4.728,
+      "eval_loss": 0.6700997352600098,
+      "eval_runtime": 97.684,
+      "eval_samples_per_second": 2.047,
+      "eval_steps_per_second": 0.512,
+      "step": 147
+    },
+    {
+      "epoch": 4.76,
+      "grad_norm": 1.6003528772082565,
+      "learning_rate": 2e-05,
+      "loss": 0.6273,
+      "step": 148
+    },
+    {
+      "epoch": 4.76,
+      "eval_loss": 0.6702569723129272,
+      "eval_runtime": 97.9207,
+      "eval_samples_per_second": 2.042,
+      "eval_steps_per_second": 0.511,
+      "step": 148
+    },
+    {
+      "epoch": 4.792,
+      "grad_norm": 1.7589414235088812,
+      "learning_rate": 2e-05,
+      "loss": 0.6506,
+      "step": 149
+    },
+    {
+      "epoch": 4.792,
+      "eval_loss": 0.6695287227630615,
+      "eval_runtime": 101.3732,
+      "eval_samples_per_second": 1.973,
+      "eval_steps_per_second": 0.493,
+      "step": 149
+    },
+    {
+      "epoch": 4.824,
+      "grad_norm": 1.6938091591501239,
+      "learning_rate": 2e-05,
+      "loss": 0.6686,
+      "step": 150
+    },
+    {
+      "epoch": 4.824,
+      "eval_loss": 0.668477475643158,
+      "eval_runtime": 97.6627,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 150
+    },
+    {
+      "epoch": 4.856,
+      "grad_norm": 1.608763198857675,
+      "learning_rate": 2e-05,
+      "loss": 0.6195,
+      "step": 151
+    },
+    {
+      "epoch": 4.856,
+      "eval_loss": 0.6693387627601624,
+      "eval_runtime": 97.7685,
+      "eval_samples_per_second": 2.046,
+      "eval_steps_per_second": 0.511,
+      "step": 151
+    },
+    {
+      "epoch": 4.888,
+      "grad_norm": 1.6097289501697254,
+      "learning_rate": 2e-05,
+      "loss": 0.6312,
+      "step": 152
+    },
+    {
+      "epoch": 4.888,
+      "eval_loss": 0.671579122543335,
+      "eval_runtime": 98.0084,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 152
+    },
+    {
+      "epoch": 4.92,
+      "grad_norm": 1.663125360797784,
+      "learning_rate": 2e-05,
+      "loss": 0.6085,
+      "step": 153
+    },
+    {
+      "epoch": 4.92,
+      "eval_loss": 0.6731156706809998,
+      "eval_runtime": 101.7316,
+      "eval_samples_per_second": 1.966,
+      "eval_steps_per_second": 0.491,
+      "step": 153
+    },
+    {
+      "epoch": 4.952,
+      "grad_norm": 1.7048758510238142,
+      "learning_rate": 2e-05,
+      "loss": 0.5892,
+      "step": 154
+    },
+    {
+      "epoch": 4.952,
+      "eval_loss": 0.6740117073059082,
+      "eval_runtime": 99.4799,
+      "eval_samples_per_second": 2.01,
+      "eval_steps_per_second": 0.503,
+      "step": 154
+    },
+    {
+      "epoch": 4.984,
+      "grad_norm": 1.76386882573663,
+      "learning_rate": 2e-05,
+      "loss": 0.672,
+      "step": 155
+    },
+    {
+      "epoch": 4.984,
+      "eval_loss": 0.6743689775466919,
+      "eval_runtime": 97.7253,
+      "eval_samples_per_second": 2.047,
+      "eval_steps_per_second": 0.512,
+      "step": 155
+    },
+    {
+      "epoch": 5.016,
+      "grad_norm": 1.7509773113974625,
+      "learning_rate": 2e-05,
+      "loss": 0.6116,
+      "step": 156
+    },
+    {
+      "epoch": 5.016,
+      "eval_loss": 0.6741358041763306,
+      "eval_runtime": 98.3883,
+      "eval_samples_per_second": 2.033,
+      "eval_steps_per_second": 0.508,
+      "step": 156
+    },
+    {
+      "epoch": 5.048,
+      "grad_norm": 1.5565965136472677,
+      "learning_rate": 2e-05,
+      "loss": 0.6127,
+      "step": 157
+    },
+    {
+      "epoch": 5.048,
+      "eval_loss": 0.6740756034851074,
+      "eval_runtime": 101.9672,
+      "eval_samples_per_second": 1.961,
+      "eval_steps_per_second": 0.49,
+      "step": 157
+    },
+    {
+      "epoch": 5.08,
+      "grad_norm": 1.9955642505530427,
+      "learning_rate": 2e-05,
+      "loss": 0.6443,
+      "step": 158
+    },
+    {
+      "epoch": 5.08,
+      "eval_loss": 0.6744682192802429,
+      "eval_runtime": 98.3076,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 158
+    },
+    {
+      "epoch": 5.112,
+      "grad_norm": 1.8176947504827694,
+      "learning_rate": 2e-05,
+      "loss": 0.5888,
+      "step": 159
+    },
+    {
+      "epoch": 5.112,
+      "eval_loss": 0.6758521795272827,
+      "eval_runtime": 98.7513,
+      "eval_samples_per_second": 2.025,
+      "eval_steps_per_second": 0.506,
+      "step": 159
+    },
+    {
+      "epoch": 5.144,
+      "grad_norm": 1.7331383222528083,
+      "learning_rate": 2e-05,
+      "loss": 0.5546,
+      "step": 160
+    },
+    {
+      "epoch": 5.144,
+      "eval_loss": 0.6774488687515259,
+      "eval_runtime": 98.3772,
+      "eval_samples_per_second": 2.033,
+      "eval_steps_per_second": 0.508,
+      "step": 160
+    },
+    {
+      "epoch": 5.176,
+      "grad_norm": 1.75241290367163,
+      "learning_rate": 2e-05,
+      "loss": 0.5998,
+      "step": 161
+    },
+    {
+      "epoch": 5.176,
+      "eval_loss": 0.6794152855873108,
+      "eval_runtime": 98.3674,
+      "eval_samples_per_second": 2.033,
+      "eval_steps_per_second": 0.508,
+      "step": 161
+    },
+    {
+      "epoch": 5.208,
+      "grad_norm": 1.7003083342136756,
+      "learning_rate": 2e-05,
+      "loss": 0.5975,
+      "step": 162
+    },
+    {
+      "epoch": 5.208,
+      "eval_loss": 0.6831446290016174,
+      "eval_runtime": 98.6795,
+      "eval_samples_per_second": 2.027,
+      "eval_steps_per_second": 0.507,
+      "step": 162
+    },
+    {
+      "epoch": 5.24,
+      "grad_norm": 1.785157318448805,
+      "learning_rate": 2e-05,
+      "loss": 0.5781,
+      "step": 163
+    },
+    {
+      "epoch": 5.24,
+      "eval_loss": 0.6875630021095276,
+      "eval_runtime": 98.4039,
+      "eval_samples_per_second": 2.032,
+      "eval_steps_per_second": 0.508,
+      "step": 163
+    },
+    {
+      "epoch": 5.272,
+      "grad_norm": 1.859036903737362,
+      "learning_rate": 2e-05,
+      "loss": 0.5615,
+      "step": 164
+    },
+    {
+      "epoch": 5.272,
+      "eval_loss": 0.693569004535675,
+      "eval_runtime": 99.2787,
+      "eval_samples_per_second": 2.015,
+      "eval_steps_per_second": 0.504,
+      "step": 164
+    },
+    {
+      "epoch": 5.304,
+      "grad_norm": 2.290233637955866,
+      "learning_rate": 2e-05,
+      "loss": 0.5671,
+      "step": 165
+    },
+    {
+      "epoch": 5.304,
+      "eval_loss": 0.6963732838630676,
+      "eval_runtime": 98.5962,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 165
+    },
+    {
+      "epoch": 5.336,
+      "grad_norm": 2.227417470514151,
+      "learning_rate": 2e-05,
+      "loss": 0.5428,
+      "step": 166
+    },
+    {
+      "epoch": 5.336,
+      "eval_loss": 0.6969068050384521,
+      "eval_runtime": 98.6958,
+      "eval_samples_per_second": 2.026,
+      "eval_steps_per_second": 0.507,
+      "step": 166
+    },
+    {
+      "epoch": 5.368,
+      "grad_norm": 1.9476792977962796,
+      "learning_rate": 2e-05,
+      "loss": 0.5565,
+      "step": 167
+    },
+    {
+      "epoch": 5.368,
+      "eval_loss": 0.6963263154029846,
+      "eval_runtime": 98.3324,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.508,
+      "step": 167
+    },
+    {
+      "epoch": 5.4,
+      "grad_norm": 2.162996766467594,
+      "learning_rate": 2e-05,
+      "loss": 0.6007,
+      "step": 168
+    },
+    {
+      "epoch": 5.4,
+      "eval_loss": 0.6939109563827515,
+      "eval_runtime": 98.4668,
+      "eval_samples_per_second": 2.031,
+      "eval_steps_per_second": 0.508,
+      "step": 168
+    },
+    {
+      "epoch": 5.432,
+      "grad_norm": 1.968981835172529,
+      "learning_rate": 2e-05,
+      "loss": 0.5998,
+      "step": 169
+    },
+    {
+      "epoch": 5.432,
+      "eval_loss": 0.6911941766738892,
+      "eval_runtime": 98.3172,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 169
+    },
+    {
+      "epoch": 5.464,
+      "grad_norm": 2.138560057900254,
+      "learning_rate": 2e-05,
+      "loss": 0.604,
+      "step": 170
+    },
+    {
+      "epoch": 5.464,
+      "eval_loss": 0.6887597441673279,
+      "eval_runtime": 98.5927,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 170
+    },
+    {
+      "epoch": 5.496,
+      "grad_norm": 2.1357688675388298,
+      "learning_rate": 2e-05,
+      "loss": 0.6759,
+      "step": 171
+    },
+    {
+      "epoch": 5.496,
+      "eval_loss": 0.6867218017578125,
+      "eval_runtime": 99.3257,
+      "eval_samples_per_second": 2.014,
+      "eval_steps_per_second": 0.503,
+      "step": 171
+    },
+    {
+      "epoch": 5.5280000000000005,
+      "grad_norm": 1.9051181997840227,
+      "learning_rate": 2e-05,
+      "loss": 0.5683,
+      "step": 172
+    },
+    {
+      "epoch": 5.5280000000000005,
+      "eval_loss": 0.6870632767677307,
+      "eval_runtime": 101.5973,
+      "eval_samples_per_second": 1.969,
+      "eval_steps_per_second": 0.492,
+      "step": 172
+    },
+    {
+      "epoch": 5.5600000000000005,
+      "grad_norm": 1.8730014322156263,
+      "learning_rate": 2e-05,
+      "loss": 0.5844,
+      "step": 173
+    },
+    {
+      "epoch": 5.5600000000000005,
+      "eval_loss": 0.687735915184021,
+      "eval_runtime": 99.0389,
+      "eval_samples_per_second": 2.019,
+      "eval_steps_per_second": 0.505,
+      "step": 173
+    },
+    {
+      "epoch": 5.592,
+      "grad_norm": 2.0975374724504303,
+      "learning_rate": 2e-05,
+      "loss": 0.5527,
+      "step": 174
+    },
+    {
+      "epoch": 5.592,
+      "eval_loss": 0.6883457899093628,
+      "eval_runtime": 103.1356,
+      "eval_samples_per_second": 1.939,
+      "eval_steps_per_second": 0.485,
+      "step": 174
+    },
+    {
+      "epoch": 5.624,
+      "grad_norm": 1.963623881621458,
+      "learning_rate": 2e-05,
+      "loss": 0.5733,
+      "step": 175
+    },
+    {
+      "epoch": 5.624,
+      "eval_loss": 0.6880955696105957,
+      "eval_runtime": 99.0308,
+      "eval_samples_per_second": 2.02,
+      "eval_steps_per_second": 0.505,
+      "step": 175
+    },
+    {
+      "epoch": 5.656,
+      "grad_norm": 1.9734678506216243,
+      "learning_rate": 2e-05,
+      "loss": 0.5736,
+      "step": 176
+    },
+    {
+      "epoch": 5.656,
+      "eval_loss": 0.6880485415458679,
+      "eval_runtime": 98.9474,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 176
+    },
+    {
+      "epoch": 5.688,
+      "grad_norm": 2.0495076030097867,
+      "learning_rate": 2e-05,
+      "loss": 0.5776,
+      "step": 177
+    },
+    {
+      "epoch": 5.688,
+      "eval_loss": 0.6872310042381287,
+      "eval_runtime": 102.5632,
+      "eval_samples_per_second": 1.95,
+      "eval_steps_per_second": 0.488,
+      "step": 177
+    },
+    {
+      "epoch": 5.72,
+      "grad_norm": 2.025156712060709,
+      "learning_rate": 2e-05,
+      "loss": 0.6222,
+      "step": 178
+    },
+    {
+      "epoch": 5.72,
+      "eval_loss": 0.6866959929466248,
+      "eval_runtime": 99.3548,
+      "eval_samples_per_second": 2.013,
+      "eval_steps_per_second": 0.503,
+      "step": 178
+    },
+    {
+      "epoch": 5.752,
+      "grad_norm": 2.0714099206695527,
+      "learning_rate": 2e-05,
+      "loss": 0.5576,
+      "step": 179
+    },
+    {
+      "epoch": 5.752,
+      "eval_loss": 0.6850792169570923,
+      "eval_runtime": 98.948,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 179
+    },
+    {
+      "epoch": 5.784,
+      "grad_norm": 2.2380767249488596,
+      "learning_rate": 2e-05,
+      "loss": 0.5628,
+      "step": 180
+    },
+    {
+      "epoch": 5.784,
+      "eval_loss": 0.6815983653068542,
+      "eval_runtime": 99.1628,
+      "eval_samples_per_second": 2.017,
+      "eval_steps_per_second": 0.504,
+      "step": 180
+    },
+    {
+      "epoch": 5.816,
+      "grad_norm": 2.104299018650581,
+      "learning_rate": 2e-05,
+      "loss": 0.6187,
+      "step": 181
+    },
+    {
+      "epoch": 5.816,
+      "eval_loss": 0.6781284809112549,
+      "eval_runtime": 99.6364,
+      "eval_samples_per_second": 2.007,
+      "eval_steps_per_second": 0.502,
+      "step": 181
+    },
+    {
+      "epoch": 5.848,
+      "grad_norm": 1.8009729510896717,
+      "learning_rate": 2e-05,
+      "loss": 0.6278,
+      "step": 182
+    },
+    {
+      "epoch": 5.848,
+      "eval_loss": 0.6774657964706421,
+      "eval_runtime": 103.8813,
+      "eval_samples_per_second": 1.925,
+      "eval_steps_per_second": 0.481,
+      "step": 182
+    },
+    {
+      "epoch": 5.88,
+      "grad_norm": 2.0126397552300004,
+      "learning_rate": 2e-05,
+      "loss": 0.5593,
+      "step": 183
+    },
+    {
+      "epoch": 5.88,
+      "eval_loss": 0.6785845756530762,
+      "eval_runtime": 99.366,
+      "eval_samples_per_second": 2.013,
+      "eval_steps_per_second": 0.503,
+      "step": 183
+    },
+    {
+      "epoch": 5.912,
+      "grad_norm": 2.0468335329470935,
+      "learning_rate": 2e-05,
+      "loss": 0.5901,
+      "step": 184
+    },
+    {
+      "epoch": 5.912,
+      "eval_loss": 0.679179847240448,
+      "eval_runtime": 99.5061,
+      "eval_samples_per_second": 2.01,
+      "eval_steps_per_second": 0.502,
+      "step": 184
+    },
+    {
+      "epoch": 5.944,
+      "grad_norm": 2.0374371712013644,
+      "learning_rate": 2e-05,
+      "loss": 0.5796,
+      "step": 185
+    },
+    {
+      "epoch": 5.944,
+      "eval_loss": 0.6800404191017151,
+      "eval_runtime": 103.6894,
+      "eval_samples_per_second": 1.929,
+      "eval_steps_per_second": 0.482,
+      "step": 185
+    },
+    {
+      "epoch": 5.976,
+      "grad_norm": 1.8830432671282609,
+      "learning_rate": 2e-05,
+      "loss": 0.5817,
+      "step": 186
+    },
+    {
+      "epoch": 5.976,
+      "eval_loss": 0.6804468631744385,
+      "eval_runtime": 99.404,
+      "eval_samples_per_second": 2.012,
+      "eval_steps_per_second": 0.503,
+      "step": 186
+    },
+    {
+      "epoch": 6.032,
+      "grad_norm": 2.017403579355368,
+      "learning_rate": 2e-05,
+      "loss": 0.5016,
+      "step": 187
+    },
+    {
+      "epoch": 6.032,
+      "eval_loss": 0.6837186217308044,
+      "eval_runtime": 100.1947,
+      "eval_samples_per_second": 1.996,
+      "eval_steps_per_second": 0.499,
+      "step": 187
+    },
+    {
+      "epoch": 6.064,
+      "grad_norm": 2.0651550543256993,
+      "learning_rate": 2e-05,
+      "loss": 0.553,
+      "step": 188
+    },
+    {
+      "epoch": 6.064,
+      "eval_loss": 0.69077467918396,
+      "eval_runtime": 95.3645,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 188
+    },
+    {
+      "epoch": 6.096,
+      "grad_norm": 1.9670810588857728,
+      "learning_rate": 2e-05,
+      "loss": 0.4774,
+      "step": 189
+    },
+    {
+      "epoch": 6.096,
+      "eval_loss": 0.6994524598121643,
+      "eval_runtime": 95.1365,
+      "eval_samples_per_second": 2.102,
+      "eval_steps_per_second": 0.526,
+      "step": 189
+    },
+    {
+      "epoch": 6.128,
+      "grad_norm": 2.3863132463229584,
+      "learning_rate": 2e-05,
+      "loss": 0.5911,
+      "step": 190
+    },
+    {
+      "epoch": 6.128,
+      "eval_loss": 0.7065584063529968,
+      "eval_runtime": 95.4502,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 190
+    },
+    {
+      "epoch": 6.16,
+      "grad_norm": 2.383412169995471,
+      "learning_rate": 2e-05,
+      "loss": 0.5147,
+      "step": 191
+    },
+    {
+      "epoch": 6.16,
+      "eval_loss": 0.7110430002212524,
+      "eval_runtime": 95.3022,
+      "eval_samples_per_second": 2.099,
+      "eval_steps_per_second": 0.525,
+      "step": 191
+    },
+    {
+      "epoch": 6.192,
+      "grad_norm": 2.232544369144536,
+      "learning_rate": 2e-05,
+      "loss": 0.5373,
+      "step": 192
+    },
+    {
+      "epoch": 6.192,
+      "eval_loss": 0.7141790986061096,
+      "eval_runtime": 95.5843,
+      "eval_samples_per_second": 2.092,
+      "eval_steps_per_second": 0.523,
+      "step": 192
+    },
+    {
+      "epoch": 6.224,
+      "grad_norm": 2.3948013626262967,
+      "learning_rate": 2e-05,
+      "loss": 0.5673,
+      "step": 193
+    },
+    {
+      "epoch": 6.224,
+      "eval_loss": 0.7145071625709534,
+      "eval_runtime": 96.0221,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 193
+    },
+    {
+      "epoch": 6.256,
+      "grad_norm": 2.282347010387328,
+      "learning_rate": 2e-05,
+      "loss": 0.4801,
+      "step": 194
+    },
+    {
+      "epoch": 6.256,
+      "eval_loss": 0.7155389189720154,
+      "eval_runtime": 95.409,
+      "eval_samples_per_second": 2.096,
+      "eval_steps_per_second": 0.524,
+      "step": 194
+    },
+    {
+      "epoch": 6.288,
+      "grad_norm": 2.7116825717896456,
+      "learning_rate": 2e-05,
+      "loss": 0.5129,
+      "step": 195
+    },
+    {
+      "epoch": 6.288,
+      "eval_loss": 0.7152263522148132,
+      "eval_runtime": 95.627,
+      "eval_samples_per_second": 2.091,
+      "eval_steps_per_second": 0.523,
+      "step": 195
+    },
+    {
+      "epoch": 6.32,
+      "grad_norm": 2.448822719551579,
+      "learning_rate": 2e-05,
+      "loss": 0.489,
+      "step": 196
+    },
+    {
+      "epoch": 6.32,
+      "eval_loss": 0.7134693264961243,
+      "eval_runtime": 96.5906,
+      "eval_samples_per_second": 2.071,
+      "eval_steps_per_second": 0.518,
+      "step": 196
+    },
+    {
+      "epoch": 6.352,
+      "grad_norm": 2.4837784920734713,
+      "learning_rate": 2e-05,
+      "loss": 0.596,
+      "step": 197
+    },
+    {
+      "epoch": 6.352,
+      "eval_loss": 0.712216317653656,
+      "eval_runtime": 96.6659,
+      "eval_samples_per_second": 2.069,
+      "eval_steps_per_second": 0.517,
+      "step": 197
+    },
+    {
+      "epoch": 6.384,
+      "grad_norm": 2.2425618673272503,
+      "learning_rate": 2e-05,
+      "loss": 0.5539,
+      "step": 198
+    },
+    {
+      "epoch": 6.384,
+      "eval_loss": 0.7105411291122437,
+      "eval_runtime": 100.6949,
+      "eval_samples_per_second": 1.986,
+      "eval_steps_per_second": 0.497,
+      "step": 198
+    },
+    {
+      "epoch": 6.416,
+      "grad_norm": 2.523025242815658,
+      "learning_rate": 2e-05,
+      "loss": 0.4946,
+      "step": 199
+    },
+    {
+      "epoch": 6.416,
+      "eval_loss": 0.7091856598854065,
+      "eval_runtime": 97.1843,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.514,
+      "step": 199
+    },
+    {
+      "epoch": 6.448,
+      "grad_norm": 2.4706535243909555,
+      "learning_rate": 2e-05,
+      "loss": 0.467,
+      "step": 200
+    },
+    {
+      "epoch": 6.448,
+      "eval_loss": 0.7090812921524048,
+      "eval_runtime": 97.069,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 200
+    },
+    {
+      "epoch": 6.48,
+      "grad_norm": 2.4466274248279833,
+      "learning_rate": 2e-05,
+      "loss": 0.55,
+      "step": 201
+    },
+    {
+      "epoch": 6.48,
+      "eval_loss": 0.7091609835624695,
+      "eval_runtime": 97.1978,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.514,
+      "step": 201
+    },
+    {
+      "epoch": 6.5120000000000005,
+      "grad_norm": 2.1567587874173717,
+      "learning_rate": 2e-05,
+      "loss": 0.5304,
+      "step": 202
+    },
+    {
+      "epoch": 6.5120000000000005,
+      "eval_loss": 0.7102614641189575,
+      "eval_runtime": 98.9298,
+      "eval_samples_per_second": 2.022,
+      "eval_steps_per_second": 0.505,
+      "step": 202
+    },
+    {
+      "epoch": 6.5440000000000005,
+      "grad_norm": 2.2160906906230533,
+      "learning_rate": 2e-05,
+      "loss": 0.5083,
+      "step": 203
+    },
+    {
+      "epoch": 6.5440000000000005,
+      "eval_loss": 0.7119375467300415,
+      "eval_runtime": 97.0494,
+      "eval_samples_per_second": 2.061,
+      "eval_steps_per_second": 0.515,
+      "step": 203
+    },
+    {
+      "epoch": 6.576,
+      "grad_norm": 2.456064686081785,
+      "learning_rate": 2e-05,
+      "loss": 0.5854,
+      "step": 204
+    },
+    {
+      "epoch": 6.576,
+      "eval_loss": 0.7119318246841431,
+      "eval_runtime": 97.165,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.515,
+      "step": 204
+    },
+    {
+      "epoch": 6.608,
+      "grad_norm": 2.5571159929171987,
+      "learning_rate": 2e-05,
+      "loss": 0.5087,
+      "step": 205
+    },
+    {
+      "epoch": 6.608,
+      "eval_loss": 0.7104336619377136,
+      "eval_runtime": 95.6354,
+      "eval_samples_per_second": 2.091,
+      "eval_steps_per_second": 0.523,
+      "step": 205
+    },
+    {
+      "epoch": 6.64,
+      "grad_norm": 2.3833463477288443,
+      "learning_rate": 2e-05,
+      "loss": 0.5928,
+      "step": 206
+    },
+    {
+      "epoch": 6.64,
+      "eval_loss": 0.7107654809951782,
+      "eval_runtime": 95.7142,
+      "eval_samples_per_second": 2.09,
+      "eval_steps_per_second": 0.522,
+      "step": 206
+    },
+    {
+      "epoch": 6.672,
+      "grad_norm": 2.6180355690739345,
+      "learning_rate": 2e-05,
+      "loss": 0.5246,
+      "step": 207
+    },
+    {
+      "epoch": 6.672,
+      "eval_loss": 0.7117599248886108,
+      "eval_runtime": 95.5237,
+      "eval_samples_per_second": 2.094,
+      "eval_steps_per_second": 0.523,
+      "step": 207
+    },
+    {
+      "epoch": 6.704,
+      "grad_norm": 2.5327556511948934,
+      "learning_rate": 2e-05,
+      "loss": 0.5142,
+      "step": 208
+    },
+    {
+      "epoch": 6.704,
+      "eval_loss": 0.7127364873886108,
+      "eval_runtime": 95.5626,
+      "eval_samples_per_second": 2.093,
+      "eval_steps_per_second": 0.523,
+      "step": 208
+    },
+    {
+      "epoch": 6.736,
+      "grad_norm": 2.7583060106689454,
+      "learning_rate": 2e-05,
+      "loss": 0.4868,
+      "step": 209
+    },
+    {
+      "epoch": 6.736,
+      "eval_loss": 0.7133345603942871,
+      "eval_runtime": 95.6791,
+      "eval_samples_per_second": 2.09,
+      "eval_steps_per_second": 0.523,
+      "step": 209
+    },
+    {
+      "epoch": 6.768,
+      "grad_norm": 2.7441146469244115,
+      "learning_rate": 2e-05,
+      "loss": 0.5935,
+      "step": 210
+    },
+    {
+      "epoch": 6.768,
+      "eval_loss": 0.7118194699287415,
+      "eval_runtime": 95.4472,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 210
+    },
+    {
+      "epoch": 6.8,
+      "grad_norm": 2.7452663949689295,
+      "learning_rate": 2e-05,
+      "loss": 0.5331,
+      "step": 211
+    },
+    {
+      "epoch": 6.8,
+      "eval_loss": 0.7106044292449951,
+      "eval_runtime": 99.8571,
+      "eval_samples_per_second": 2.003,
+      "eval_steps_per_second": 0.501,
+      "step": 211
+    },
+    {
+      "epoch": 6.832,
+      "grad_norm": 2.454028115196185,
+      "learning_rate": 2e-05,
+      "loss": 0.5491,
+      "step": 212
+    },
+    {
+      "epoch": 6.832,
+      "eval_loss": 0.7110446691513062,
+      "eval_runtime": 96.9239,
+      "eval_samples_per_second": 2.063,
+      "eval_steps_per_second": 0.516,
+      "step": 212
+    },
+    {
+      "epoch": 6.864,
+      "grad_norm": 2.758969331776205,
+      "learning_rate": 2e-05,
+      "loss": 0.5442,
+      "step": 213
+    },
+    {
+      "epoch": 6.864,
+      "eval_loss": 0.710289478302002,
+      "eval_runtime": 96.7481,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 213
+    },
+    {
+      "epoch": 6.896,
+      "grad_norm": 2.5394726231273457,
+      "learning_rate": 2e-05,
+      "loss": 0.5422,
+      "step": 214
+    },
+    {
+      "epoch": 6.896,
+      "eval_loss": 0.7084080576896667,
+      "eval_runtime": 98.6547,
+      "eval_samples_per_second": 2.027,
+      "eval_steps_per_second": 0.507,
+      "step": 214
+    },
+    {
+      "epoch": 6.928,
+      "grad_norm": 2.642919051796586,
+      "learning_rate": 2e-05,
+      "loss": 0.5721,
+      "step": 215
+    },
+    {
+      "epoch": 6.928,
+      "eval_loss": 0.7063835263252258,
+      "eval_runtime": 96.9686,
+      "eval_samples_per_second": 2.063,
+      "eval_steps_per_second": 0.516,
+      "step": 215
+    },
+    {
+      "epoch": 6.96,
+      "grad_norm": 2.501395694239305,
+      "learning_rate": 2e-05,
+      "loss": 0.5742,
+      "step": 216
+    },
+    {
+      "epoch": 6.96,
+      "eval_loss": 0.7037466168403625,
+      "eval_runtime": 96.5477,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 216
+    },
+    {
+      "epoch": 6.992,
+      "grad_norm": 2.685414325564133,
+      "learning_rate": 2e-05,
+      "loss": 0.5195,
+      "step": 217
+    },
+    {
+      "epoch": 6.992,
+      "eval_loss": 0.7025836110115051,
+      "eval_runtime": 97.0823,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 217
+    },
+    {
+      "epoch": 7.024,
+      "grad_norm": 2.558047543529016,
+      "learning_rate": 2e-05,
+      "loss": 0.4689,
+      "step": 218
+    },
+    {
+      "epoch": 7.024,
+      "eval_loss": 0.7044098377227783,
+      "eval_runtime": 96.8828,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 218
+    },
+    {
+      "epoch": 7.056,
+      "grad_norm": 2.4415187474081192,
+      "learning_rate": 2e-05,
+      "loss": 0.4898,
+      "step": 219
+    },
+    {
+      "epoch": 7.056,
+      "eval_loss": 0.7104077935218811,
+      "eval_runtime": 96.9482,
+      "eval_samples_per_second": 2.063,
+      "eval_steps_per_second": 0.516,
+      "step": 219
+    },
+    {
+      "epoch": 7.088,
+      "grad_norm": 2.4475533939573153,
+      "learning_rate": 2e-05,
+      "loss": 0.4652,
+      "step": 220
+    },
+    {
+      "epoch": 7.088,
+      "eval_loss": 0.7213549613952637,
+      "eval_runtime": 96.8958,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 220
+    },
+    {
+      "epoch": 7.12,
+      "grad_norm": 2.604960196075638,
+      "learning_rate": 2e-05,
+      "loss": 0.5096,
+      "step": 221
+    },
+    {
+      "epoch": 7.12,
+      "eval_loss": 0.7343713641166687,
+      "eval_runtime": 97.2891,
+      "eval_samples_per_second": 2.056,
+      "eval_steps_per_second": 0.514,
+      "step": 221
+    },
+    {
+      "epoch": 7.152,
+      "grad_norm": 3.1059901537521135,
+      "learning_rate": 2e-05,
+      "loss": 0.4678,
+      "step": 222
+    },
+    {
+      "epoch": 7.152,
+      "eval_loss": 0.7466915845870972,
+      "eval_runtime": 97.7786,
+      "eval_samples_per_second": 2.045,
+      "eval_steps_per_second": 0.511,
+      "step": 222
+    },
+    {
+      "epoch": 7.184,
+      "grad_norm": 2.6643302536817,
+      "learning_rate": 2e-05,
+      "loss": 0.4807,
+      "step": 223
+    },
+    {
+      "epoch": 7.184,
+      "eval_loss": 0.7558013796806335,
+      "eval_runtime": 97.7416,
+      "eval_samples_per_second": 2.046,
+      "eval_steps_per_second": 0.512,
+      "step": 223
+    },
+    {
+      "epoch": 7.216,
+      "grad_norm": 3.218895177437451,
+      "learning_rate": 2e-05,
+      "loss": 0.4556,
+      "step": 224
+    },
+    {
+      "epoch": 7.216,
+      "eval_loss": 0.7590299844741821,
+      "eval_runtime": 97.7479,
+      "eval_samples_per_second": 2.046,
+      "eval_steps_per_second": 0.512,
+      "step": 224
+    },
+    {
+      "epoch": 7.248,
+      "grad_norm": 3.367892941528959,
+      "learning_rate": 2e-05,
+      "loss": 0.4572,
+      "step": 225
+    },
+    {
+      "epoch": 7.248,
+      "eval_loss": 0.7567412853240967,
+      "eval_runtime": 97.6242,
+      "eval_samples_per_second": 2.049,
+      "eval_steps_per_second": 0.512,
+      "step": 225
+    },
+    {
+      "epoch": 7.28,
+      "grad_norm": 2.860324389327654,
+      "learning_rate": 2e-05,
+      "loss": 0.4989,
+      "step": 226
+    },
+    {
+      "epoch": 7.28,
+      "eval_loss": 0.7538604736328125,
+      "eval_runtime": 97.8342,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 226
+    },
+    {
+      "epoch": 7.312,
+      "grad_norm": 2.783566742283421,
+      "learning_rate": 2e-05,
+      "loss": 0.5011,
+      "step": 227
+    },
+    {
+      "epoch": 7.312,
+      "eval_loss": 0.7509406805038452,
+      "eval_runtime": 98.0668,
+      "eval_samples_per_second": 2.039,
+      "eval_steps_per_second": 0.51,
+      "step": 227
+    },
+    {
+      "epoch": 7.344,
+      "grad_norm": 2.95622825050015,
+      "learning_rate": 2e-05,
+      "loss": 0.4632,
+      "step": 228
+    },
+    {
+      "epoch": 7.344,
+      "eval_loss": 0.7484157085418701,
+      "eval_runtime": 97.8003,
+      "eval_samples_per_second": 2.045,
+      "eval_steps_per_second": 0.511,
+      "step": 228
+    },
+    {
+      "epoch": 7.376,
+      "grad_norm": 2.868982236682872,
+      "learning_rate": 2e-05,
+      "loss": 0.5131,
+      "step": 229
+    },
+    {
+      "epoch": 7.376,
+      "eval_loss": 0.7442774772644043,
+      "eval_runtime": 97.779,
+      "eval_samples_per_second": 2.045,
+      "eval_steps_per_second": 0.511,
+      "step": 229
+    },
+    {
+      "epoch": 7.408,
+      "grad_norm": 2.8398125186341456,
+      "learning_rate": 2e-05,
+      "loss": 0.4589,
+      "step": 230
+    },
+    {
+      "epoch": 7.408,
+      "eval_loss": 0.7437787055969238,
+      "eval_runtime": 100.3989,
+      "eval_samples_per_second": 1.992,
+      "eval_steps_per_second": 0.498,
+      "step": 230
+    },
+    {
+      "epoch": 7.44,
+      "grad_norm": 2.860835801821795,
+      "learning_rate": 2e-05,
+      "loss": 0.4551,
+      "step": 231
+    },
+    {
+      "epoch": 7.44,
+      "eval_loss": 0.7474150061607361,
+      "eval_runtime": 100.3908,
+      "eval_samples_per_second": 1.992,
+      "eval_steps_per_second": 0.498,
+      "step": 231
+    },
+    {
+      "epoch": 7.4719999999999995,
+      "grad_norm": 3.061812810287673,
+      "learning_rate": 2e-05,
+      "loss": 0.4393,
+      "step": 232
+    },
+    {
+      "epoch": 7.4719999999999995,
+      "eval_loss": 0.751905083656311,
+      "eval_runtime": 98.0564,
+      "eval_samples_per_second": 2.04,
+      "eval_steps_per_second": 0.51,
+      "step": 232
+    },
+    {
+      "epoch": 7.504,
+      "grad_norm": 2.9460244368130764,
+      "learning_rate": 2e-05,
+      "loss": 0.4885,
+      "step": 233
+    },
+    {
+      "epoch": 7.504,
+      "eval_loss": 0.7555700540542603,
+      "eval_runtime": 97.6649,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 233
+    },
+    {
+      "epoch": 7.536,
+      "grad_norm": 3.195386368100557,
+      "learning_rate": 2e-05,
+      "loss": 0.4825,
+      "step": 234
+    },
+    {
+      "epoch": 7.536,
+      "eval_loss": 0.7574471831321716,
+      "eval_runtime": 97.8436,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 234
+    },
+    {
+      "epoch": 7.568,
+      "grad_norm": 3.2267422014122853,
+      "learning_rate": 2e-05,
+      "loss": 0.4369,
+      "step": 235
+    },
+    {
+      "epoch": 7.568,
+      "eval_loss": 0.7590253949165344,
+      "eval_runtime": 97.8616,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 235
+    },
+    {
+      "epoch": 7.608,
+      "grad_norm": 3.0309471686984115,
+      "learning_rate": 2e-05,
+      "loss": 0.4533,
+      "step": 236
+    },
+    {
+      "epoch": 7.608,
+      "eval_loss": 0.7582225203514099,
+      "eval_runtime": 102.1646,
+      "eval_samples_per_second": 1.958,
+      "eval_steps_per_second": 0.489,
+      "step": 236
+    },
+    {
+      "epoch": 7.64,
+      "grad_norm": 3.0386242391155456,
+      "learning_rate": 2e-05,
+      "loss": 0.4557,
+      "step": 237
+    },
+    {
+      "epoch": 7.64,
+      "eval_loss": 0.757209062576294,
+      "eval_runtime": 96.2842,
+      "eval_samples_per_second": 2.077,
+      "eval_steps_per_second": 0.519,
+      "step": 237
+    },
+    {
+      "epoch": 7.672,
+      "grad_norm": 3.595657373206686,
+      "learning_rate": 2e-05,
+      "loss": 0.493,
+      "step": 238
+    },
+    {
+      "epoch": 7.672,
+      "eval_loss": 0.7517231702804565,
+      "eval_runtime": 95.4225,
+      "eval_samples_per_second": 2.096,
+      "eval_steps_per_second": 0.524,
+      "step": 238
+    },
+    {
+      "epoch": 7.704,
+      "grad_norm": 3.246444664818878,
+      "learning_rate": 2e-05,
+      "loss": 0.4663,
+      "step": 239
+    },
+    {
+      "epoch": 7.704,
+      "eval_loss": 0.7478169202804565,
+      "eval_runtime": 95.3771,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 239
+    },
+    {
+      "epoch": 7.736,
+      "grad_norm": 2.9436367134600827,
+      "learning_rate": 2e-05,
+      "loss": 0.4871,
+      "step": 240
+    },
+    {
+      "epoch": 7.736,
+      "eval_loss": 0.7445845603942871,
+      "eval_runtime": 98.2477,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 240
+    },
+    {
+      "epoch": 7.768,
+      "grad_norm": 2.8534339094113284,
+      "learning_rate": 2e-05,
+      "loss": 0.5307,
+      "step": 241
+    },
+    {
+      "epoch": 7.768,
+      "eval_loss": 0.7433913946151733,
+      "eval_runtime": 97.3211,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 241
+    },
+    {
+      "epoch": 7.8,
+      "grad_norm": 3.180683551713395,
+      "learning_rate": 2e-05,
+      "loss": 0.4189,
+      "step": 242
+    },
+    {
+      "epoch": 7.8,
+      "eval_loss": 0.7443408370018005,
+      "eval_runtime": 97.045,
+      "eval_samples_per_second": 2.061,
+      "eval_steps_per_second": 0.515,
+      "step": 242
+    },
+    {
+      "epoch": 7.832,
+      "grad_norm": 2.915485597080642,
+      "learning_rate": 2e-05,
+      "loss": 0.466,
+      "step": 243
+    },
+    {
+      "epoch": 7.832,
+      "eval_loss": 0.7447941303253174,
+      "eval_runtime": 97.3217,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 243
+    },
+    {
+      "epoch": 7.864,
+      "grad_norm": 3.116167020951512,
+      "learning_rate": 2e-05,
+      "loss": 0.5012,
+      "step": 244
+    },
+    {
+      "epoch": 7.864,
+      "eval_loss": 0.744431734085083,
+      "eval_runtime": 97.2911,
+      "eval_samples_per_second": 2.056,
+      "eval_steps_per_second": 0.514,
+      "step": 244
+    },
+    {
+      "epoch": 7.896,
+      "grad_norm": 3.0808644763492694,
+      "learning_rate": 2e-05,
+      "loss": 0.4809,
+      "step": 245
+    },
+    {
+      "epoch": 7.896,
+      "eval_loss": 0.7447794079780579,
+      "eval_runtime": 100.9572,
+      "eval_samples_per_second": 1.981,
+      "eval_steps_per_second": 0.495,
+      "step": 245
+    },
+    {
+      "epoch": 7.928,
+      "grad_norm": 3.0956122448213836,
+      "learning_rate": 2e-05,
+      "loss": 0.4685,
+      "step": 246
+    },
+    {
+      "epoch": 7.928,
+      "eval_loss": 0.7456656694412231,
+      "eval_runtime": 99.4752,
+      "eval_samples_per_second": 2.011,
+      "eval_steps_per_second": 0.503,
+      "step": 246
+    },
+    {
+      "epoch": 7.96,
+      "grad_norm": 3.1578427063687804,
+      "learning_rate": 2e-05,
+      "loss": 0.5019,
+      "step": 247
+    },
+    {
+      "epoch": 7.96,
+      "eval_loss": 0.7482015490531921,
+      "eval_runtime": 100.5527,
+      "eval_samples_per_second": 1.989,
+      "eval_steps_per_second": 0.497,
+      "step": 247
+    },
+    {
+      "epoch": 7.992,
+      "grad_norm": 3.0384545980818096,
+      "learning_rate": 2e-05,
+      "loss": 0.4667,
+      "step": 248
+    },
+    {
+      "epoch": 7.992,
+      "eval_loss": 0.7531342506408691,
+      "eval_runtime": 96.9014,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 248
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 248,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 8,
+  "save_steps": 5,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 215944051097600.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}

checkpoint-248/training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fb674b9c15f7660edd7f2ca5081ae35468486e4f7de4435b1dcf26181111b74d
+size 8184

checkpoint-248/zero_to_fp32.py

@@ -0,0 +1,604 @@
+#!/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, exclude_frozen_parameters):
+    """
+    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,
+                                                          exclude_frozen_parameters)
+    elif zero_stage == 3:
+        return _get_fp32_state_dict_from_zero3_checkpoint(world_size, fp32_flat_groups, zero_model_states,
+                                                          exclude_frozen_parameters)
+
+
+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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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,
+                                               exclude_frozen_parameters):
+    state_dict = OrderedDict()
+
+    # buffers
+    buffers = zero_model_states[0].buffers
+    state_dict.update(buffers)
+    if debug:
+        print(f"added {len(buffers)} buffers")
+
+    if not exclude_frozen_parameters:
+        _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, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+
+    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, exclude_frozen_parameters)
+
+
+def convert_zero_checkpoint_to_fp32_state_dict(checkpoint_dir, output_file, tag=None, exclude_frozen_parameters=False):
+    """
+    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``
+        - ``exclude_frozen_parameters``: exclude frozen parameters
+    """
+
+    state_dict = get_fp32_state_dict_from_zero_checkpoint(checkpoint_dir, tag, exclude_frozen_parameters)
+    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("--exclude_frozen_parameters", action='store_true', help="exclude frozen parameters")
+    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,
+                                               exclude_frozen_parameters=args.exclude_frozen_parameters)

config.json

@@ -0,0 +1,75 @@
+{
+  "_name_or_path": "liuhaotian/llava-v1.6-mistral-7b",
+  "architectures": [
+    "LlavaMistralForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 1,
+  "eos_token_id": 2,
+  "freeze_mm_mlp_adapter": false,
+  "freeze_mm_vision_resampler": false,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "image_aspect_ratio": "anyres",
+  "image_crop_resolution": 224,
+  "image_grid_pinpoints": [
+    [
+      336,
+      672
+    ],
+    [
+      672,
+      336
+    ],
+    [
+      672,
+      672
+    ],
+    [
+      1008,
+      336
+    ],
+    [
+      336,
+      1008
+    ]
+  ],
+  "image_split_resolution": 224,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 32768,
+  "mlp_bias": false,
+  "mm_hidden_size": 1024,
+  "mm_patch_merge_type": "spatial_unpad",
+  "mm_projector_lr": 2e-05,
+  "mm_projector_type": "mlp2x_gelu",
+  "mm_resampler_type": null,
+  "mm_use_im_patch_token": false,
+  "mm_use_im_start_end": false,
+  "mm_vision_select_feature": "patch",
+  "mm_vision_select_layer": -2,
+  "mm_vision_tower": "openai/clip-vit-large-patch14-336",
+  "mm_vision_tower_lr": 2e-06,
+  "model_type": "llava_llama",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "pretraining_tp": 1,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 1000000.0,
+  "sliding_window": null,
+  "tie_word_embeddings": false,
+  "tokenizer_model_max_length": 4096,
+  "tokenizer_padding_side": "right",
+  "torch_dtype": "bfloat16",
+  "transformers_version": "4.45.2",
+  "tune_mm_mlp_adapter": false,
+  "tune_mm_vision_resampler": false,
+  "unfreeze_mm_vision_tower": true,
+  "use_cache": true,
+  "use_mm_proj": true,
+  "vocab_size": 32000
+}

non_lora_trainables.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f56da5e7651600f211c76f543351b71b7ec5119e0d7ee19bcf199ba7e1e40d1c
+size 41961648

optimizer.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:135305bb6bf98c6a50c9369d9fb2d05d8eb578e6c55df1eb604eddc6f87d56fc
+size 258787874

trainer_state.json

@@ -0,0 +1,2367 @@
+{
+  "best_metric": 0.6666322350502014,
+  "best_model_checkpoint": "./checkpoints/llava-v1.6-mistral/checkpoint-125",
+  "epoch": 4.992,
+  "eval_steps": 1.0,
+  "global_step": 155,
+  "is_hyper_param_search": false,
+  "is_local_process_zero": true,
+  "is_world_process_zero": true,
+  "log_history": [
+    {
+      "epoch": 0.032,
+      "grad_norm": 1.3562940564009667,
+      "learning_rate": 0.0,
+      "loss": 1.2069,
+      "step": 1
+    },
+    {
+      "epoch": 0.032,
+      "eval_loss": 1.189530849456787,
+      "eval_runtime": 100.6384,
+      "eval_samples_per_second": 1.987,
+      "eval_steps_per_second": 0.497,
+      "step": 1
+    },
+    {
+      "epoch": 0.064,
+      "grad_norm": 1.3726335882966116,
+      "learning_rate": 7.737056144690831e-06,
+      "loss": 1.242,
+      "step": 2
+    },
+    {
+      "epoch": 0.064,
+      "eval_loss": 1.189530849456787,
+      "eval_runtime": 95.3713,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 2
+    },
+    {
+      "epoch": 0.096,
+      "grad_norm": 1.267071497789874,
+      "learning_rate": 1.2262943855309169e-05,
+      "loss": 1.1588,
+      "step": 3
+    },
+    {
+      "epoch": 0.096,
+      "eval_loss": 1.1675384044647217,
+      "eval_runtime": 99.4768,
+      "eval_samples_per_second": 2.011,
+      "eval_steps_per_second": 0.503,
+      "step": 3
+    },
+    {
+      "epoch": 0.128,
+      "grad_norm": 1.2066781722006759,
+      "learning_rate": 1.5474112289381662e-05,
+      "loss": 1.1196,
+      "step": 4
+    },
+    {
+      "epoch": 0.128,
+      "eval_loss": 1.1391011476516724,
+      "eval_runtime": 95.4777,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 4
+    },
+    {
+      "epoch": 0.16,
+      "grad_norm": 1.2746210039866817,
+      "learning_rate": 1.7964888034078543e-05,
+      "loss": 1.1462,
+      "step": 5
+    },
+    {
+      "epoch": 0.16,
+      "eval_loss": 1.1023523807525635,
+      "eval_runtime": 95.4138,
+      "eval_samples_per_second": 2.096,
+      "eval_steps_per_second": 0.524,
+      "step": 5
+    },
+    {
+      "epoch": 0.192,
+      "grad_norm": 1.0552650848120237,
+      "learning_rate": 1.9999999999999998e-05,
+      "loss": 1.1147,
+      "step": 6
+    },
+    {
+      "epoch": 0.192,
+      "eval_loss": 1.063745379447937,
+      "eval_runtime": 97.6633,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 6
+    },
+    {
+      "epoch": 0.224,
+      "grad_norm": 1.0779287740182022,
+      "learning_rate": 2e-05,
+      "loss": 1.1773,
+      "step": 7
+    },
+    {
+      "epoch": 0.224,
+      "eval_loss": 1.024451732635498,
+      "eval_runtime": 97.388,
+      "eval_samples_per_second": 2.054,
+      "eval_steps_per_second": 0.513,
+      "step": 7
+    },
+    {
+      "epoch": 0.256,
+      "grad_norm": 0.9918200495785524,
+      "learning_rate": 2e-05,
+      "loss": 1.0434,
+      "step": 8
+    },
+    {
+      "epoch": 0.256,
+      "eval_loss": 0.9906812906265259,
+      "eval_runtime": 97.2423,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 8
+    },
+    {
+      "epoch": 0.288,
+      "grad_norm": 0.9119239308208719,
+      "learning_rate": 2e-05,
+      "loss": 1.0527,
+      "step": 9
+    },
+    {
+      "epoch": 0.288,
+      "eval_loss": 0.9626247882843018,
+      "eval_runtime": 97.2048,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.514,
+      "step": 9
+    },
+    {
+      "epoch": 0.32,
+      "grad_norm": 0.7292073828818868,
+      "learning_rate": 2e-05,
+      "loss": 0.9801,
+      "step": 10
+    },
+    {
+      "epoch": 0.32,
+      "eval_loss": 0.9422003030776978,
+      "eval_runtime": 97.143,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 10
+    },
+    {
+      "epoch": 0.352,
+      "grad_norm": 0.8195137858461904,
+      "learning_rate": 2e-05,
+      "loss": 1.0104,
+      "step": 11
+    },
+    {
+      "epoch": 0.352,
+      "eval_loss": 0.9264165759086609,
+      "eval_runtime": 97.2279,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 11
+    },
+    {
+      "epoch": 0.384,
+      "grad_norm": 0.7060464565699565,
+      "learning_rate": 2e-05,
+      "loss": 0.9764,
+      "step": 12
+    },
+    {
+      "epoch": 0.384,
+      "eval_loss": 0.9138455986976624,
+      "eval_runtime": 97.0954,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 12
+    },
+    {
+      "epoch": 0.416,
+      "grad_norm": 0.8207130544099034,
+      "learning_rate": 2e-05,
+      "loss": 1.0371,
+      "step": 13
+    },
+    {
+      "epoch": 0.416,
+      "eval_loss": 0.902658224105835,
+      "eval_runtime": 100.8924,
+      "eval_samples_per_second": 1.982,
+      "eval_steps_per_second": 0.496,
+      "step": 13
+    },
+    {
+      "epoch": 0.448,
+      "grad_norm": 0.7962544586919155,
+      "learning_rate": 2e-05,
+      "loss": 0.9455,
+      "step": 14
+    },
+    {
+      "epoch": 0.448,
+      "eval_loss": 0.8926536440849304,
+      "eval_runtime": 100.856,
+      "eval_samples_per_second": 1.983,
+      "eval_steps_per_second": 0.496,
+      "step": 14
+    },
+    {
+      "epoch": 0.48,
+      "grad_norm": 0.8246421268957655,
+      "learning_rate": 2e-05,
+      "loss": 0.9456,
+      "step": 15
+    },
+    {
+      "epoch": 0.48,
+      "eval_loss": 0.8823295831680298,
+      "eval_runtime": 97.0568,
+      "eval_samples_per_second": 2.061,
+      "eval_steps_per_second": 0.515,
+      "step": 15
+    },
+    {
+      "epoch": 0.512,
+      "grad_norm": 0.7040845257818263,
+      "learning_rate": 2e-05,
+      "loss": 0.9283,
+      "step": 16
+    },
+    {
+      "epoch": 0.512,
+      "eval_loss": 0.8721897602081299,
+      "eval_runtime": 98.318,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 16
+    },
+    {
+      "epoch": 0.544,
+      "grad_norm": 0.7904707927869743,
+      "learning_rate": 2e-05,
+      "loss": 0.9786,
+      "step": 17
+    },
+    {
+      "epoch": 0.544,
+      "eval_loss": 0.861838698387146,
+      "eval_runtime": 101.4487,
+      "eval_samples_per_second": 1.971,
+      "eval_steps_per_second": 0.493,
+      "step": 17
+    },
+    {
+      "epoch": 0.576,
+      "grad_norm": 0.7605583546638608,
+      "learning_rate": 2e-05,
+      "loss": 0.9361,
+      "step": 18
+    },
+    {
+      "epoch": 0.576,
+      "eval_loss": 0.8519415259361267,
+      "eval_runtime": 98.2448,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 18
+    },
+    {
+      "epoch": 0.608,
+      "grad_norm": 0.7005232039249588,
+      "learning_rate": 2e-05,
+      "loss": 0.8707,
+      "step": 19
+    },
+    {
+      "epoch": 0.608,
+      "eval_loss": 0.842995822429657,
+      "eval_runtime": 98.3165,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 19
+    },
+    {
+      "epoch": 0.64,
+      "grad_norm": 0.653693259495668,
+      "learning_rate": 2e-05,
+      "loss": 0.9332,
+      "step": 20
+    },
+    {
+      "epoch": 0.64,
+      "eval_loss": 0.8355565071105957,
+      "eval_runtime": 98.237,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 20
+    },
+    {
+      "epoch": 0.672,
+      "grad_norm": 0.6600451404133434,
+      "learning_rate": 2e-05,
+      "loss": 0.8848,
+      "step": 21
+    },
+    {
+      "epoch": 0.672,
+      "eval_loss": 0.8297473788261414,
+      "eval_runtime": 98.1825,
+      "eval_samples_per_second": 2.037,
+      "eval_steps_per_second": 0.509,
+      "step": 21
+    },
+    {
+      "epoch": 0.704,
+      "grad_norm": 0.6666525650519819,
+      "learning_rate": 2e-05,
+      "loss": 0.9261,
+      "step": 22
+    },
+    {
+      "epoch": 0.704,
+      "eval_loss": 0.8248497247695923,
+      "eval_runtime": 98.1541,
+      "eval_samples_per_second": 2.038,
+      "eval_steps_per_second": 0.509,
+      "step": 22
+    },
+    {
+      "epoch": 0.736,
+      "grad_norm": 0.6759718697778233,
+      "learning_rate": 2e-05,
+      "loss": 1.0103,
+      "step": 23
+    },
+    {
+      "epoch": 0.736,
+      "eval_loss": 0.8214980959892273,
+      "eval_runtime": 98.2494,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 23
+    },
+    {
+      "epoch": 0.768,
+      "grad_norm": 0.651870418762904,
+      "learning_rate": 2e-05,
+      "loss": 0.9538,
+      "step": 24
+    },
+    {
+      "epoch": 0.768,
+      "eval_loss": 0.818403959274292,
+      "eval_runtime": 97.9476,
+      "eval_samples_per_second": 2.042,
+      "eval_steps_per_second": 0.51,
+      "step": 24
+    },
+    {
+      "epoch": 0.8,
+      "grad_norm": 0.6985872283132858,
+      "learning_rate": 2e-05,
+      "loss": 0.8216,
+      "step": 25
+    },
+    {
+      "epoch": 0.8,
+      "eval_loss": 0.8149588108062744,
+      "eval_runtime": 95.6175,
+      "eval_samples_per_second": 2.092,
+      "eval_steps_per_second": 0.523,
+      "step": 25
+    },
+    {
+      "epoch": 0.832,
+      "grad_norm": 0.7801182369335715,
+      "learning_rate": 2e-05,
+      "loss": 0.8478,
+      "step": 26
+    },
+    {
+      "epoch": 0.832,
+      "eval_loss": 0.8100517392158508,
+      "eval_runtime": 99.7733,
+      "eval_samples_per_second": 2.005,
+      "eval_steps_per_second": 0.501,
+      "step": 26
+    },
+    {
+      "epoch": 0.864,
+      "grad_norm": 0.6727766374526198,
+      "learning_rate": 2e-05,
+      "loss": 0.9015,
+      "step": 27
+    },
+    {
+      "epoch": 0.864,
+      "eval_loss": 0.8051578998565674,
+      "eval_runtime": 95.7387,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 27
+    },
+    {
+      "epoch": 0.896,
+      "grad_norm": 0.7398381040653764,
+      "learning_rate": 2e-05,
+      "loss": 0.8434,
+      "step": 28
+    },
+    {
+      "epoch": 0.896,
+      "eval_loss": 0.7998965382575989,
+      "eval_runtime": 95.7753,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 28
+    },
+    {
+      "epoch": 0.928,
+      "grad_norm": 0.6837341164482282,
+      "learning_rate": 2e-05,
+      "loss": 0.9124,
+      "step": 29
+    },
+    {
+      "epoch": 0.928,
+      "eval_loss": 0.7946869134902954,
+      "eval_runtime": 95.933,
+      "eval_samples_per_second": 2.085,
+      "eval_steps_per_second": 0.521,
+      "step": 29
+    },
+    {
+      "epoch": 0.96,
+      "grad_norm": 0.7285508052728015,
+      "learning_rate": 2e-05,
+      "loss": 0.8252,
+      "step": 30
+    },
+    {
+      "epoch": 0.96,
+      "eval_loss": 0.7891057729721069,
+      "eval_runtime": 95.871,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.522,
+      "step": 30
+    },
+    {
+      "epoch": 0.992,
+      "grad_norm": 0.706524788728722,
+      "learning_rate": 2e-05,
+      "loss": 0.8382,
+      "step": 31
+    },
+    {
+      "epoch": 0.992,
+      "eval_loss": 0.7835636138916016,
+      "eval_runtime": 95.7491,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 31
+    },
+    {
+      "epoch": 1.024,
+      "grad_norm": 0.7392763969473979,
+      "learning_rate": 2e-05,
+      "loss": 0.8586,
+      "step": 32
+    },
+    {
+      "epoch": 1.024,
+      "eval_loss": 0.7780101299285889,
+      "eval_runtime": 99.7493,
+      "eval_samples_per_second": 2.005,
+      "eval_steps_per_second": 0.501,
+      "step": 32
+    },
+    {
+      "epoch": 1.056,
+      "grad_norm": 0.7407143277408553,
+      "learning_rate": 2e-05,
+      "loss": 0.8438,
+      "step": 33
+    },
+    {
+      "epoch": 1.056,
+      "eval_loss": 0.7728690505027771,
+      "eval_runtime": 95.8863,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.521,
+      "step": 33
+    },
+    {
+      "epoch": 1.088,
+      "grad_norm": 0.7146296362894146,
+      "learning_rate": 2e-05,
+      "loss": 0.8357,
+      "step": 34
+    },
+    {
+      "epoch": 1.088,
+      "eval_loss": 0.7683935761451721,
+      "eval_runtime": 97.824,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 34
+    },
+    {
+      "epoch": 1.12,
+      "grad_norm": 0.7117142678212836,
+      "learning_rate": 2e-05,
+      "loss": 0.7884,
+      "step": 35
+    },
+    {
+      "epoch": 1.12,
+      "eval_loss": 0.7646369338035583,
+      "eval_runtime": 95.8185,
+      "eval_samples_per_second": 2.087,
+      "eval_steps_per_second": 0.522,
+      "step": 35
+    },
+    {
+      "epoch": 1.152,
+      "grad_norm": 0.7735767817076163,
+      "learning_rate": 2e-05,
+      "loss": 0.8851,
+      "step": 36
+    },
+    {
+      "epoch": 1.152,
+      "eval_loss": 0.7616337537765503,
+      "eval_runtime": 96.1525,
+      "eval_samples_per_second": 2.08,
+      "eval_steps_per_second": 0.52,
+      "step": 36
+    },
+    {
+      "epoch": 1.184,
+      "grad_norm": 0.7386952203362822,
+      "learning_rate": 2e-05,
+      "loss": 0.7943,
+      "step": 37
+    },
+    {
+      "epoch": 1.184,
+      "eval_loss": 0.759408712387085,
+      "eval_runtime": 96.2772,
+      "eval_samples_per_second": 2.077,
+      "eval_steps_per_second": 0.519,
+      "step": 37
+    },
+    {
+      "epoch": 1.216,
+      "grad_norm": 0.6688305832985106,
+      "learning_rate": 2e-05,
+      "loss": 0.8433,
+      "step": 38
+    },
+    {
+      "epoch": 1.216,
+      "eval_loss": 0.7577520608901978,
+      "eval_runtime": 95.7726,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 38
+    },
+    {
+      "epoch": 1.248,
+      "grad_norm": 0.6861808214947546,
+      "learning_rate": 2e-05,
+      "loss": 0.8132,
+      "step": 39
+    },
+    {
+      "epoch": 1.248,
+      "eval_loss": 0.7565059065818787,
+      "eval_runtime": 96.1577,
+      "eval_samples_per_second": 2.08,
+      "eval_steps_per_second": 0.52,
+      "step": 39
+    },
+    {
+      "epoch": 1.28,
+      "grad_norm": 0.7450886292190618,
+      "learning_rate": 2e-05,
+      "loss": 0.9067,
+      "step": 40
+    },
+    {
+      "epoch": 1.28,
+      "eval_loss": 0.7562046647071838,
+      "eval_runtime": 95.9983,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 40
+    },
+    {
+      "epoch": 1.312,
+      "grad_norm": 0.7508586498301053,
+      "learning_rate": 2e-05,
+      "loss": 0.8642,
+      "step": 41
+    },
+    {
+      "epoch": 1.312,
+      "eval_loss": 0.755351722240448,
+      "eval_runtime": 96.6244,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 41
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "grad_norm": 0.7717273184776977,
+      "learning_rate": 2e-05,
+      "loss": 0.9406,
+      "step": 42
+    },
+    {
+      "epoch": 1.3439999999999999,
+      "eval_loss": 0.753773033618927,
+      "eval_runtime": 100.8127,
+      "eval_samples_per_second": 1.984,
+      "eval_steps_per_second": 0.496,
+      "step": 42
+    },
+    {
+      "epoch": 1.376,
+      "grad_norm": 0.7980115302909253,
+      "learning_rate": 2e-05,
+      "loss": 0.7732,
+      "step": 43
+    },
+    {
+      "epoch": 1.376,
+      "eval_loss": 0.7511720061302185,
+      "eval_runtime": 97.4372,
+      "eval_samples_per_second": 2.053,
+      "eval_steps_per_second": 0.513,
+      "step": 43
+    },
+    {
+      "epoch": 1.408,
+      "grad_norm": 0.794554904521861,
+      "learning_rate": 2e-05,
+      "loss": 0.8962,
+      "step": 44
+    },
+    {
+      "epoch": 1.408,
+      "eval_loss": 0.7484354376792908,
+      "eval_runtime": 96.7621,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 44
+    },
+    {
+      "epoch": 1.44,
+      "grad_norm": 0.7964935137573818,
+      "learning_rate": 2e-05,
+      "loss": 0.7725,
+      "step": 45
+    },
+    {
+      "epoch": 1.44,
+      "eval_loss": 0.7454251050949097,
+      "eval_runtime": 96.5292,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 45
+    },
+    {
+      "epoch": 1.472,
+      "grad_norm": 0.7477437094752549,
+      "learning_rate": 2e-05,
+      "loss": 0.8238,
+      "step": 46
+    },
+    {
+      "epoch": 1.472,
+      "eval_loss": 0.7427694201469421,
+      "eval_runtime": 96.4788,
+      "eval_samples_per_second": 2.073,
+      "eval_steps_per_second": 0.518,
+      "step": 46
+    },
+    {
+      "epoch": 1.504,
+      "grad_norm": 0.7443339980176984,
+      "learning_rate": 2e-05,
+      "loss": 0.8273,
+      "step": 47
+    },
+    {
+      "epoch": 1.504,
+      "eval_loss": 0.7407447099685669,
+      "eval_runtime": 96.5686,
+      "eval_samples_per_second": 2.071,
+      "eval_steps_per_second": 0.518,
+      "step": 47
+    },
+    {
+      "epoch": 1.536,
+      "grad_norm": 0.7901404920428264,
+      "learning_rate": 2e-05,
+      "loss": 0.7974,
+      "step": 48
+    },
+    {
+      "epoch": 1.536,
+      "eval_loss": 0.7381842136383057,
+      "eval_runtime": 96.6071,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.518,
+      "step": 48
+    },
+    {
+      "epoch": 1.568,
+      "grad_norm": 0.8677109521747711,
+      "learning_rate": 2e-05,
+      "loss": 0.8011,
+      "step": 49
+    },
+    {
+      "epoch": 1.568,
+      "eval_loss": 0.73555588722229,
+      "eval_runtime": 96.4189,
+      "eval_samples_per_second": 2.074,
+      "eval_steps_per_second": 0.519,
+      "step": 49
+    },
+    {
+      "epoch": 1.6,
+      "grad_norm": 0.8170436579363425,
+      "learning_rate": 2e-05,
+      "loss": 0.7762,
+      "step": 50
+    },
+    {
+      "epoch": 1.6,
+      "eval_loss": 0.7331439256668091,
+      "eval_runtime": 100.5875,
+      "eval_samples_per_second": 1.988,
+      "eval_steps_per_second": 0.497,
+      "step": 50
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "grad_norm": 0.8190539924095326,
+      "learning_rate": 2e-05,
+      "loss": 0.7469,
+      "step": 51
+    },
+    {
+      "epoch": 1.6320000000000001,
+      "eval_loss": 0.7306647300720215,
+      "eval_runtime": 101.1429,
+      "eval_samples_per_second": 1.977,
+      "eval_steps_per_second": 0.494,
+      "step": 51
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "grad_norm": 0.8044932603192197,
+      "learning_rate": 2e-05,
+      "loss": 0.8052,
+      "step": 52
+    },
+    {
+      "epoch": 1.6640000000000001,
+      "eval_loss": 0.7287429571151733,
+      "eval_runtime": 97.25,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 52
+    },
+    {
+      "epoch": 1.696,
+      "grad_norm": 0.7811170846407103,
+      "learning_rate": 2e-05,
+      "loss": 0.755,
+      "step": 53
+    },
+    {
+      "epoch": 1.696,
+      "eval_loss": 0.7264651656150818,
+      "eval_runtime": 97.67,
+      "eval_samples_per_second": 2.048,
+      "eval_steps_per_second": 0.512,
+      "step": 53
+    },
+    {
+      "epoch": 1.728,
+      "grad_norm": 0.808405093976468,
+      "learning_rate": 2e-05,
+      "loss": 0.7838,
+      "step": 54
+    },
+    {
+      "epoch": 1.728,
+      "eval_loss": 0.7242828607559204,
+      "eval_runtime": 101.2008,
+      "eval_samples_per_second": 1.976,
+      "eval_steps_per_second": 0.494,
+      "step": 54
+    },
+    {
+      "epoch": 1.76,
+      "grad_norm": 0.8172106270954758,
+      "learning_rate": 2e-05,
+      "loss": 0.7123,
+      "step": 55
+    },
+    {
+      "epoch": 1.76,
+      "eval_loss": 0.7217574119567871,
+      "eval_runtime": 99.6145,
+      "eval_samples_per_second": 2.008,
+      "eval_steps_per_second": 0.502,
+      "step": 55
+    },
+    {
+      "epoch": 1.792,
+      "grad_norm": 0.83212530473105,
+      "learning_rate": 2e-05,
+      "loss": 0.8141,
+      "step": 56
+    },
+    {
+      "epoch": 1.792,
+      "eval_loss": 0.7192932367324829,
+      "eval_runtime": 97.1377,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 56
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "grad_norm": 0.8632612959683078,
+      "learning_rate": 2e-05,
+      "loss": 0.8005,
+      "step": 57
+    },
+    {
+      "epoch": 1.8239999999999998,
+      "eval_loss": 0.717003345489502,
+      "eval_runtime": 97.2871,
+      "eval_samples_per_second": 2.056,
+      "eval_steps_per_second": 0.514,
+      "step": 57
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "grad_norm": 0.8362081570841255,
+      "learning_rate": 2e-05,
+      "loss": 0.8254,
+      "step": 58
+    },
+    {
+      "epoch": 1.8559999999999999,
+      "eval_loss": 0.715446949005127,
+      "eval_runtime": 97.5751,
+      "eval_samples_per_second": 2.05,
+      "eval_steps_per_second": 0.512,
+      "step": 58
+    },
+    {
+      "epoch": 1.888,
+      "grad_norm": 0.8862779841020042,
+      "learning_rate": 2e-05,
+      "loss": 0.7949,
+      "step": 59
+    },
+    {
+      "epoch": 1.888,
+      "eval_loss": 0.7145370244979858,
+      "eval_runtime": 97.2406,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 59
+    },
+    {
+      "epoch": 1.92,
+      "grad_norm": 0.875656256488838,
+      "learning_rate": 2e-05,
+      "loss": 0.8277,
+      "step": 60
+    },
+    {
+      "epoch": 1.92,
+      "eval_loss": 0.7140547037124634,
+      "eval_runtime": 97.1485,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 60
+    },
+    {
+      "epoch": 1.952,
+      "grad_norm": 0.8850690645543244,
+      "learning_rate": 2e-05,
+      "loss": 0.8117,
+      "step": 61
+    },
+    {
+      "epoch": 1.952,
+      "eval_loss": 0.7128701210021973,
+      "eval_runtime": 97.8906,
+      "eval_samples_per_second": 2.043,
+      "eval_steps_per_second": 0.511,
+      "step": 61
+    },
+    {
+      "epoch": 1.984,
+      "grad_norm": 0.8961641824408927,
+      "learning_rate": 2e-05,
+      "loss": 0.704,
+      "step": 62
+    },
+    {
+      "epoch": 1.984,
+      "eval_loss": 0.7107064127922058,
+      "eval_runtime": 102.2163,
+      "eval_samples_per_second": 1.957,
+      "eval_steps_per_second": 0.489,
+      "step": 62
+    },
+    {
+      "epoch": 2.016,
+      "grad_norm": 0.8602247372004996,
+      "learning_rate": 2e-05,
+      "loss": 0.8016,
+      "step": 63
+    },
+    {
+      "epoch": 2.016,
+      "eval_loss": 0.7091581225395203,
+      "eval_runtime": 98.2114,
+      "eval_samples_per_second": 2.036,
+      "eval_steps_per_second": 0.509,
+      "step": 63
+    },
+    {
+      "epoch": 2.048,
+      "grad_norm": 0.7881703384753784,
+      "learning_rate": 2e-05,
+      "loss": 0.7912,
+      "step": 64
+    },
+    {
+      "epoch": 2.048,
+      "eval_loss": 0.7081906199455261,
+      "eval_runtime": 98.9403,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 64
+    },
+    {
+      "epoch": 2.08,
+      "grad_norm": 0.8436680506716614,
+      "learning_rate": 2e-05,
+      "loss": 0.6965,
+      "step": 65
+    },
+    {
+      "epoch": 2.08,
+      "eval_loss": 0.7070262432098389,
+      "eval_runtime": 97.7451,
+      "eval_samples_per_second": 2.046,
+      "eval_steps_per_second": 0.512,
+      "step": 65
+    },
+    {
+      "epoch": 2.112,
+      "grad_norm": 0.8694446846234115,
+      "learning_rate": 2e-05,
+      "loss": 0.7015,
+      "step": 66
+    },
+    {
+      "epoch": 2.112,
+      "eval_loss": 0.7055197358131409,
+      "eval_runtime": 97.994,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 66
+    },
+    {
+      "epoch": 2.144,
+      "grad_norm": 0.9518475022730614,
+      "learning_rate": 2e-05,
+      "loss": 0.6952,
+      "step": 67
+    },
+    {
+      "epoch": 2.144,
+      "eval_loss": 0.7035844326019287,
+      "eval_runtime": 98.0296,
+      "eval_samples_per_second": 2.04,
+      "eval_steps_per_second": 0.51,
+      "step": 67
+    },
+    {
+      "epoch": 2.176,
+      "grad_norm": 0.8662066024104106,
+      "learning_rate": 2e-05,
+      "loss": 0.7731,
+      "step": 68
+    },
+    {
+      "epoch": 2.176,
+      "eval_loss": 0.7019283771514893,
+      "eval_runtime": 100.0982,
+      "eval_samples_per_second": 1.998,
+      "eval_steps_per_second": 0.5,
+      "step": 68
+    },
+    {
+      "epoch": 2.208,
+      "grad_norm": 0.9574631942368209,
+      "learning_rate": 2e-05,
+      "loss": 0.7572,
+      "step": 69
+    },
+    {
+      "epoch": 2.208,
+      "eval_loss": 0.7011401653289795,
+      "eval_runtime": 101.9267,
+      "eval_samples_per_second": 1.962,
+      "eval_steps_per_second": 0.491,
+      "step": 69
+    },
+    {
+      "epoch": 2.24,
+      "grad_norm": 0.9477542521472675,
+      "learning_rate": 2e-05,
+      "loss": 0.7393,
+      "step": 70
+    },
+    {
+      "epoch": 2.24,
+      "eval_loss": 0.7006180286407471,
+      "eval_runtime": 101.9672,
+      "eval_samples_per_second": 1.961,
+      "eval_steps_per_second": 0.49,
+      "step": 70
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "grad_norm": 0.9612956116350272,
+      "learning_rate": 2e-05,
+      "loss": 0.7175,
+      "step": 71
+    },
+    {
+      "epoch": 2.2720000000000002,
+      "eval_loss": 0.7002539038658142,
+      "eval_runtime": 101.8755,
+      "eval_samples_per_second": 1.963,
+      "eval_steps_per_second": 0.491,
+      "step": 71
+    },
+    {
+      "epoch": 2.304,
+      "grad_norm": 0.9934844303955727,
+      "learning_rate": 2e-05,
+      "loss": 0.791,
+      "step": 72
+    },
+    {
+      "epoch": 2.304,
+      "eval_loss": 0.6997203230857849,
+      "eval_runtime": 97.9926,
+      "eval_samples_per_second": 2.041,
+      "eval_steps_per_second": 0.51,
+      "step": 72
+    },
+    {
+      "epoch": 2.336,
+      "grad_norm": 1.0540430431227044,
+      "learning_rate": 2e-05,
+      "loss": 0.7542,
+      "step": 73
+    },
+    {
+      "epoch": 2.336,
+      "eval_loss": 0.6988361477851868,
+      "eval_runtime": 100.3704,
+      "eval_samples_per_second": 1.993,
+      "eval_steps_per_second": 0.498,
+      "step": 73
+    },
+    {
+      "epoch": 2.368,
+      "grad_norm": 1.0249397957961794,
+      "learning_rate": 2e-05,
+      "loss": 0.8084,
+      "step": 74
+    },
+    {
+      "epoch": 2.368,
+      "eval_loss": 0.6980065703392029,
+      "eval_runtime": 101.8958,
+      "eval_samples_per_second": 1.963,
+      "eval_steps_per_second": 0.491,
+      "step": 74
+    },
+    {
+      "epoch": 2.4,
+      "grad_norm": 1.0445498365690145,
+      "learning_rate": 2e-05,
+      "loss": 0.7964,
+      "step": 75
+    },
+    {
+      "epoch": 2.4,
+      "eval_loss": 0.6971798539161682,
+      "eval_runtime": 98.1624,
+      "eval_samples_per_second": 2.037,
+      "eval_steps_per_second": 0.509,
+      "step": 75
+    },
+    {
+      "epoch": 2.432,
+      "grad_norm": 0.9685893079320761,
+      "learning_rate": 2e-05,
+      "loss": 0.8047,
+      "step": 76
+    },
+    {
+      "epoch": 2.432,
+      "eval_loss": 0.696861743927002,
+      "eval_runtime": 98.6404,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 76
+    },
+    {
+      "epoch": 2.464,
+      "grad_norm": 0.9753091933204456,
+      "learning_rate": 2e-05,
+      "loss": 0.684,
+      "step": 77
+    },
+    {
+      "epoch": 2.464,
+      "eval_loss": 0.6957904100418091,
+      "eval_runtime": 98.8766,
+      "eval_samples_per_second": 2.023,
+      "eval_steps_per_second": 0.506,
+      "step": 77
+    },
+    {
+      "epoch": 2.496,
+      "grad_norm": 0.9389149478503764,
+      "learning_rate": 2e-05,
+      "loss": 0.7434,
+      "step": 78
+    },
+    {
+      "epoch": 2.496,
+      "eval_loss": 0.6943306922912598,
+      "eval_runtime": 98.5751,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 78
+    },
+    {
+      "epoch": 2.528,
+      "grad_norm": 1.0675154845211299,
+      "learning_rate": 2e-05,
+      "loss": 0.7208,
+      "step": 79
+    },
+    {
+      "epoch": 2.528,
+      "eval_loss": 0.6920651197433472,
+      "eval_runtime": 99.0851,
+      "eval_samples_per_second": 2.018,
+      "eval_steps_per_second": 0.505,
+      "step": 79
+    },
+    {
+      "epoch": 2.56,
+      "grad_norm": 0.9937936593307737,
+      "learning_rate": 2e-05,
+      "loss": 0.6948,
+      "step": 80
+    },
+    {
+      "epoch": 2.56,
+      "eval_loss": 0.6899142265319824,
+      "eval_runtime": 98.7974,
+      "eval_samples_per_second": 2.024,
+      "eval_steps_per_second": 0.506,
+      "step": 80
+    },
+    {
+      "epoch": 2.592,
+      "grad_norm": 0.9650832276698476,
+      "learning_rate": 2e-05,
+      "loss": 0.7666,
+      "step": 81
+    },
+    {
+      "epoch": 2.592,
+      "eval_loss": 0.6886695623397827,
+      "eval_runtime": 98.6813,
+      "eval_samples_per_second": 2.027,
+      "eval_steps_per_second": 0.507,
+      "step": 81
+    },
+    {
+      "epoch": 2.624,
+      "grad_norm": 0.9961610958296112,
+      "learning_rate": 2e-05,
+      "loss": 0.7741,
+      "step": 82
+    },
+    {
+      "epoch": 2.624,
+      "eval_loss": 0.687745213508606,
+      "eval_runtime": 98.624,
+      "eval_samples_per_second": 2.028,
+      "eval_steps_per_second": 0.507,
+      "step": 82
+    },
+    {
+      "epoch": 2.656,
+      "grad_norm": 1.0261499699526089,
+      "learning_rate": 2e-05,
+      "loss": 0.7747,
+      "step": 83
+    },
+    {
+      "epoch": 2.656,
+      "eval_loss": 0.6869972944259644,
+      "eval_runtime": 98.9604,
+      "eval_samples_per_second": 2.021,
+      "eval_steps_per_second": 0.505,
+      "step": 83
+    },
+    {
+      "epoch": 2.6879999999999997,
+      "grad_norm": 1.025059237745532,
+      "learning_rate": 2e-05,
+      "loss": 0.7526,
+      "step": 84
+    },
+    {
+      "epoch": 2.6879999999999997,
+      "eval_loss": 0.6862147450447083,
+      "eval_runtime": 98.4387,
+      "eval_samples_per_second": 2.032,
+      "eval_steps_per_second": 0.508,
+      "step": 84
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "grad_norm": 1.1383626266013125,
+      "learning_rate": 2e-05,
+      "loss": 0.683,
+      "step": 85
+    },
+    {
+      "epoch": 2.7199999999999998,
+      "eval_loss": 0.6845572590827942,
+      "eval_runtime": 98.4979,
+      "eval_samples_per_second": 2.03,
+      "eval_steps_per_second": 0.508,
+      "step": 85
+    },
+    {
+      "epoch": 2.752,
+      "grad_norm": 1.0427739679421295,
+      "learning_rate": 2e-05,
+      "loss": 0.7362,
+      "step": 86
+    },
+    {
+      "epoch": 2.752,
+      "eval_loss": 0.683280348777771,
+      "eval_runtime": 103.6707,
+      "eval_samples_per_second": 1.929,
+      "eval_steps_per_second": 0.482,
+      "step": 86
+    },
+    {
+      "epoch": 2.784,
+      "grad_norm": 1.1280129141879938,
+      "learning_rate": 2e-05,
+      "loss": 0.7743,
+      "step": 87
+    },
+    {
+      "epoch": 2.784,
+      "eval_loss": 0.6824235320091248,
+      "eval_runtime": 99.2995,
+      "eval_samples_per_second": 2.014,
+      "eval_steps_per_second": 0.504,
+      "step": 87
+    },
+    {
+      "epoch": 2.816,
+      "grad_norm": 1.0017715497784696,
+      "learning_rate": 2e-05,
+      "loss": 0.7164,
+      "step": 88
+    },
+    {
+      "epoch": 2.816,
+      "eval_loss": 0.6819013357162476,
+      "eval_runtime": 99.5206,
+      "eval_samples_per_second": 2.01,
+      "eval_steps_per_second": 0.502,
+      "step": 88
+    },
+    {
+      "epoch": 2.848,
+      "grad_norm": 1.0769673642284994,
+      "learning_rate": 2e-05,
+      "loss": 0.7013,
+      "step": 89
+    },
+    {
+      "epoch": 2.848,
+      "eval_loss": 0.6820746064186096,
+      "eval_runtime": 96.5197,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 89
+    },
+    {
+      "epoch": 2.88,
+      "grad_norm": 1.0389167281844591,
+      "learning_rate": 2e-05,
+      "loss": 0.6805,
+      "step": 90
+    },
+    {
+      "epoch": 2.88,
+      "eval_loss": 0.6826525926589966,
+      "eval_runtime": 96.0942,
+      "eval_samples_per_second": 2.081,
+      "eval_steps_per_second": 0.52,
+      "step": 90
+    },
+    {
+      "epoch": 2.912,
+      "grad_norm": 0.9705743838620626,
+      "learning_rate": 2e-05,
+      "loss": 0.7681,
+      "step": 91
+    },
+    {
+      "epoch": 2.912,
+      "eval_loss": 0.6836435198783875,
+      "eval_runtime": 96.2441,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 91
+    },
+    {
+      "epoch": 2.944,
+      "grad_norm": 1.037823791993831,
+      "learning_rate": 2e-05,
+      "loss": 0.7532,
+      "step": 92
+    },
+    {
+      "epoch": 2.944,
+      "eval_loss": 0.6845746040344238,
+      "eval_runtime": 96.2394,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 92
+    },
+    {
+      "epoch": 2.976,
+      "grad_norm": 1.1323835942146157,
+      "learning_rate": 2e-05,
+      "loss": 0.7171,
+      "step": 93
+    },
+    {
+      "epoch": 2.976,
+      "eval_loss": 0.684663712978363,
+      "eval_runtime": 96.23,
+      "eval_samples_per_second": 2.078,
+      "eval_steps_per_second": 0.52,
+      "step": 93
+    },
+    {
+      "epoch": 3.008,
+      "grad_norm": 1.1957864756602699,
+      "learning_rate": 2e-05,
+      "loss": 0.7166,
+      "step": 94
+    },
+    {
+      "epoch": 3.008,
+      "eval_loss": 0.6830846667289734,
+      "eval_runtime": 96.6549,
+      "eval_samples_per_second": 2.069,
+      "eval_steps_per_second": 0.517,
+      "step": 94
+    },
+    {
+      "epoch": 3.04,
+      "grad_norm": 1.1077357794232636,
+      "learning_rate": 2e-05,
+      "loss": 0.6667,
+      "step": 95
+    },
+    {
+      "epoch": 3.04,
+      "eval_loss": 0.6810076832771301,
+      "eval_runtime": 96.3239,
+      "eval_samples_per_second": 2.076,
+      "eval_steps_per_second": 0.519,
+      "step": 95
+    },
+    {
+      "epoch": 3.072,
+      "grad_norm": 1.1851219157184936,
+      "learning_rate": 2e-05,
+      "loss": 0.7115,
+      "step": 96
+    },
+    {
+      "epoch": 3.072,
+      "eval_loss": 0.6796395778656006,
+      "eval_runtime": 96.9109,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 96
+    },
+    {
+      "epoch": 3.104,
+      "grad_norm": 1.0812671042616444,
+      "learning_rate": 2e-05,
+      "loss": 0.7333,
+      "step": 97
+    },
+    {
+      "epoch": 3.104,
+      "eval_loss": 0.6794567108154297,
+      "eval_runtime": 96.7403,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 97
+    },
+    {
+      "epoch": 3.136,
+      "grad_norm": 1.130095597839828,
+      "learning_rate": 2e-05,
+      "loss": 0.6328,
+      "step": 98
+    },
+    {
+      "epoch": 3.136,
+      "eval_loss": 0.6792007684707642,
+      "eval_runtime": 96.8136,
+      "eval_samples_per_second": 2.066,
+      "eval_steps_per_second": 0.516,
+      "step": 98
+    },
+    {
+      "epoch": 3.168,
+      "grad_norm": 1.16102100344116,
+      "learning_rate": 2e-05,
+      "loss": 0.6625,
+      "step": 99
+    },
+    {
+      "epoch": 3.168,
+      "eval_loss": 0.6789132952690125,
+      "eval_runtime": 96.6982,
+      "eval_samples_per_second": 2.068,
+      "eval_steps_per_second": 0.517,
+      "step": 99
+    },
+    {
+      "epoch": 3.2,
+      "grad_norm": 1.226689811951201,
+      "learning_rate": 2e-05,
+      "loss": 0.7522,
+      "step": 100
+    },
+    {
+      "epoch": 3.2,
+      "eval_loss": 0.6786602735519409,
+      "eval_runtime": 96.6253,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 100
+    },
+    {
+      "epoch": 3.232,
+      "grad_norm": 1.1623462595850367,
+      "learning_rate": 2e-05,
+      "loss": 0.6755,
+      "step": 101
+    },
+    {
+      "epoch": 3.232,
+      "eval_loss": 0.6791322827339172,
+      "eval_runtime": 96.5222,
+      "eval_samples_per_second": 2.072,
+      "eval_steps_per_second": 0.518,
+      "step": 101
+    },
+    {
+      "epoch": 3.2640000000000002,
+      "grad_norm": 1.16303930181089,
+      "learning_rate": 2e-05,
+      "loss": 0.6752,
+      "step": 102
+    },
+    {
+      "epoch": 3.2640000000000002,
+      "eval_loss": 0.680637538433075,
+      "eval_runtime": 96.8789,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 102
+    },
+    {
+      "epoch": 3.296,
+      "grad_norm": 1.166028046661615,
+      "learning_rate": 2e-05,
+      "loss": 0.6732,
+      "step": 103
+    },
+    {
+      "epoch": 3.296,
+      "eval_loss": 0.6818951964378357,
+      "eval_runtime": 96.7579,
+      "eval_samples_per_second": 2.067,
+      "eval_steps_per_second": 0.517,
+      "step": 103
+    },
+    {
+      "epoch": 3.328,
+      "grad_norm": 1.2872370423601793,
+      "learning_rate": 2e-05,
+      "loss": 0.7713,
+      "step": 104
+    },
+    {
+      "epoch": 3.328,
+      "eval_loss": 0.682328462600708,
+      "eval_runtime": 98.3142,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 104
+    },
+    {
+      "epoch": 3.36,
+      "grad_norm": 1.1363822202896854,
+      "learning_rate": 2e-05,
+      "loss": 0.7429,
+      "step": 105
+    },
+    {
+      "epoch": 3.36,
+      "eval_loss": 0.6817943453788757,
+      "eval_runtime": 100.7507,
+      "eval_samples_per_second": 1.985,
+      "eval_steps_per_second": 0.496,
+      "step": 105
+    },
+    {
+      "epoch": 3.416,
+      "grad_norm": 1.2711044658075554,
+      "learning_rate": 2e-05,
+      "loss": 0.7057,
+      "step": 106
+    },
+    {
+      "epoch": 3.416,
+      "eval_loss": 0.6794378161430359,
+      "eval_runtime": 100.4096,
+      "eval_samples_per_second": 1.992,
+      "eval_steps_per_second": 0.498,
+      "step": 106
+    },
+    {
+      "epoch": 3.448,
+      "grad_norm": 1.3212295597772596,
+      "learning_rate": 2e-05,
+      "loss": 0.6982,
+      "step": 107
+    },
+    {
+      "epoch": 3.448,
+      "eval_loss": 0.6767404079437256,
+      "eval_runtime": 96.6386,
+      "eval_samples_per_second": 2.07,
+      "eval_steps_per_second": 0.517,
+      "step": 107
+    },
+    {
+      "epoch": 3.48,
+      "grad_norm": 1.2420948881737728,
+      "learning_rate": 2e-05,
+      "loss": 0.7092,
+      "step": 108
+    },
+    {
+      "epoch": 3.48,
+      "eval_loss": 0.6751566529273987,
+      "eval_runtime": 95.1851,
+      "eval_samples_per_second": 2.101,
+      "eval_steps_per_second": 0.525,
+      "step": 108
+    },
+    {
+      "epoch": 3.512,
+      "grad_norm": 1.2965352636029341,
+      "learning_rate": 2e-05,
+      "loss": 0.6715,
+      "step": 109
+    },
+    {
+      "epoch": 3.512,
+      "eval_loss": 0.6750080585479736,
+      "eval_runtime": 95.3479,
+      "eval_samples_per_second": 2.098,
+      "eval_steps_per_second": 0.524,
+      "step": 109
+    },
+    {
+      "epoch": 3.544,
+      "grad_norm": 1.2789534479099607,
+      "learning_rate": 2e-05,
+      "loss": 0.6732,
+      "step": 110
+    },
+    {
+      "epoch": 3.544,
+      "eval_loss": 0.6744334697723389,
+      "eval_runtime": 95.3541,
+      "eval_samples_per_second": 2.097,
+      "eval_steps_per_second": 0.524,
+      "step": 110
+    },
+    {
+      "epoch": 3.576,
+      "grad_norm": 1.3881035995379567,
+      "learning_rate": 2e-05,
+      "loss": 0.6777,
+      "step": 111
+    },
+    {
+      "epoch": 3.576,
+      "eval_loss": 0.6730498671531677,
+      "eval_runtime": 96.972,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.516,
+      "step": 111
+    },
+    {
+      "epoch": 3.608,
+      "grad_norm": 1.2398078245019133,
+      "learning_rate": 2e-05,
+      "loss": 0.6314,
+      "step": 112
+    },
+    {
+      "epoch": 3.608,
+      "eval_loss": 0.6725335717201233,
+      "eval_runtime": 97.1794,
+      "eval_samples_per_second": 2.058,
+      "eval_steps_per_second": 0.515,
+      "step": 112
+    },
+    {
+      "epoch": 3.64,
+      "grad_norm": 1.3383993031041075,
+      "learning_rate": 2e-05,
+      "loss": 0.6776,
+      "step": 113
+    },
+    {
+      "epoch": 3.64,
+      "eval_loss": 0.6719880104064941,
+      "eval_runtime": 97.2319,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 113
+    },
+    {
+      "epoch": 3.672,
+      "grad_norm": 1.289557205987285,
+      "learning_rate": 2e-05,
+      "loss": 0.7475,
+      "step": 114
+    },
+    {
+      "epoch": 3.672,
+      "eval_loss": 0.6723533868789673,
+      "eval_runtime": 97.3108,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 114
+    },
+    {
+      "epoch": 3.7039999999999997,
+      "grad_norm": 1.121644844950096,
+      "learning_rate": 2e-05,
+      "loss": 0.6813,
+      "step": 115
+    },
+    {
+      "epoch": 3.7039999999999997,
+      "eval_loss": 0.673932671546936,
+      "eval_runtime": 97.0919,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 115
+    },
+    {
+      "epoch": 3.7359999999999998,
+      "grad_norm": 1.5370551636574723,
+      "learning_rate": 2e-05,
+      "loss": 0.7036,
+      "step": 116
+    },
+    {
+      "epoch": 3.7359999999999998,
+      "eval_loss": 0.6737978458404541,
+      "eval_runtime": 97.0064,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.515,
+      "step": 116
+    },
+    {
+      "epoch": 3.768,
+      "grad_norm": 1.3017720066449985,
+      "learning_rate": 2e-05,
+      "loss": 0.7048,
+      "step": 117
+    },
+    {
+      "epoch": 3.768,
+      "eval_loss": 0.6731936931610107,
+      "eval_runtime": 96.9134,
+      "eval_samples_per_second": 2.064,
+      "eval_steps_per_second": 0.516,
+      "step": 117
+    },
+    {
+      "epoch": 3.8,
+      "grad_norm": 1.3976157986974596,
+      "learning_rate": 2e-05,
+      "loss": 0.6512,
+      "step": 118
+    },
+    {
+      "epoch": 3.8,
+      "eval_loss": 0.6716210842132568,
+      "eval_runtime": 97.4028,
+      "eval_samples_per_second": 2.053,
+      "eval_steps_per_second": 0.513,
+      "step": 118
+    },
+    {
+      "epoch": 3.832,
+      "grad_norm": 1.3982632064891347,
+      "learning_rate": 2e-05,
+      "loss": 0.6481,
+      "step": 119
+    },
+    {
+      "epoch": 3.832,
+      "eval_loss": 0.6690347194671631,
+      "eval_runtime": 97.2096,
+      "eval_samples_per_second": 2.057,
+      "eval_steps_per_second": 0.514,
+      "step": 119
+    },
+    {
+      "epoch": 3.864,
+      "grad_norm": 1.346077433712507,
+      "learning_rate": 2e-05,
+      "loss": 0.6005,
+      "step": 120
+    },
+    {
+      "epoch": 3.864,
+      "eval_loss": 0.667186975479126,
+      "eval_runtime": 97.326,
+      "eval_samples_per_second": 2.055,
+      "eval_steps_per_second": 0.514,
+      "step": 120
+    },
+    {
+      "epoch": 3.896,
+      "grad_norm": 1.3142652448176726,
+      "learning_rate": 2e-05,
+      "loss": 0.6967,
+      "step": 121
+    },
+    {
+      "epoch": 3.896,
+      "eval_loss": 0.6662881970405579,
+      "eval_runtime": 101.7648,
+      "eval_samples_per_second": 1.965,
+      "eval_steps_per_second": 0.491,
+      "step": 121
+    },
+    {
+      "epoch": 3.928,
+      "grad_norm": 1.3368669812290557,
+      "learning_rate": 2e-05,
+      "loss": 0.6954,
+      "step": 122
+    },
+    {
+      "epoch": 3.928,
+      "eval_loss": 0.6659301519393921,
+      "eval_runtime": 98.3123,
+      "eval_samples_per_second": 2.034,
+      "eval_steps_per_second": 0.509,
+      "step": 122
+    },
+    {
+      "epoch": 3.96,
+      "grad_norm": 1.3992228513792793,
+      "learning_rate": 2e-05,
+      "loss": 0.6168,
+      "step": 123
+    },
+    {
+      "epoch": 3.96,
+      "eval_loss": 0.6662861704826355,
+      "eval_runtime": 98.5527,
+      "eval_samples_per_second": 2.029,
+      "eval_steps_per_second": 0.507,
+      "step": 123
+    },
+    {
+      "epoch": 3.992,
+      "grad_norm": 1.403303492690172,
+      "learning_rate": 2e-05,
+      "loss": 0.6818,
+      "step": 124
+    },
+    {
+      "epoch": 3.992,
+      "eval_loss": 0.666582465171814,
+      "eval_runtime": 95.7576,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 124
+    },
+    {
+      "epoch": 4.024,
+      "grad_norm": 1.37383602466541,
+      "learning_rate": 2e-05,
+      "loss": 0.6539,
+      "step": 125
+    },
+    {
+      "epoch": 4.024,
+      "eval_loss": 0.6666322350502014,
+      "eval_runtime": 95.5774,
+      "eval_samples_per_second": 2.093,
+      "eval_steps_per_second": 0.523,
+      "step": 125
+    },
+    {
+      "epoch": 4.064,
+      "grad_norm": 1.3710348437226973,
+      "learning_rate": 2e-05,
+      "loss": 0.7181,
+      "step": 126
+    },
+    {
+      "epoch": 4.064,
+      "eval_loss": 0.6673327684402466,
+      "eval_runtime": 102.0166,
+      "eval_samples_per_second": 1.96,
+      "eval_steps_per_second": 0.49,
+      "step": 126
+    },
+    {
+      "epoch": 4.096,
+      "grad_norm": 1.297767286241014,
+      "learning_rate": 2e-05,
+      "loss": 0.6452,
+      "step": 127
+    },
+    {
+      "epoch": 4.096,
+      "eval_loss": 0.6687185764312744,
+      "eval_runtime": 95.6988,
+      "eval_samples_per_second": 2.09,
+      "eval_steps_per_second": 0.522,
+      "step": 127
+    },
+    {
+      "epoch": 4.128,
+      "grad_norm": 1.292569885504323,
+      "learning_rate": 2e-05,
+      "loss": 0.6145,
+      "step": 128
+    },
+    {
+      "epoch": 4.128,
+      "eval_loss": 0.6715303063392639,
+      "eval_runtime": 95.465,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 128
+    },
+    {
+      "epoch": 4.16,
+      "grad_norm": 1.419162300463038,
+      "learning_rate": 2e-05,
+      "loss": 0.6088,
+      "step": 129
+    },
+    {
+      "epoch": 4.16,
+      "eval_loss": 0.675749659538269,
+      "eval_runtime": 97.8829,
+      "eval_samples_per_second": 2.043,
+      "eval_steps_per_second": 0.511,
+      "step": 129
+    },
+    {
+      "epoch": 4.192,
+      "grad_norm": 1.5215794897159234,
+      "learning_rate": 2e-05,
+      "loss": 0.6391,
+      "step": 130
+    },
+    {
+      "epoch": 4.192,
+      "eval_loss": 0.677629828453064,
+      "eval_runtime": 95.4624,
+      "eval_samples_per_second": 2.095,
+      "eval_steps_per_second": 0.524,
+      "step": 130
+    },
+    {
+      "epoch": 4.224,
+      "grad_norm": 1.4684908212096284,
+      "learning_rate": 2e-05,
+      "loss": 0.6102,
+      "step": 131
+    },
+    {
+      "epoch": 4.224,
+      "eval_loss": 0.6779782176017761,
+      "eval_runtime": 97.1518,
+      "eval_samples_per_second": 2.059,
+      "eval_steps_per_second": 0.515,
+      "step": 131
+    },
+    {
+      "epoch": 4.256,
+      "grad_norm": 1.7036283256760842,
+      "learning_rate": 2e-05,
+      "loss": 0.6396,
+      "step": 132
+    },
+    {
+      "epoch": 4.256,
+      "eval_loss": 0.6759344339370728,
+      "eval_runtime": 97.8632,
+      "eval_samples_per_second": 2.044,
+      "eval_steps_per_second": 0.511,
+      "step": 132
+    },
+    {
+      "epoch": 4.288,
+      "grad_norm": 1.6051118551476093,
+      "learning_rate": 2e-05,
+      "loss": 0.6566,
+      "step": 133
+    },
+    {
+      "epoch": 4.288,
+      "eval_loss": 0.673258364200592,
+      "eval_runtime": 97.4448,
+      "eval_samples_per_second": 2.052,
+      "eval_steps_per_second": 0.513,
+      "step": 133
+    },
+    {
+      "epoch": 4.32,
+      "grad_norm": 1.5572448859529642,
+      "learning_rate": 2e-05,
+      "loss": 0.6223,
+      "step": 134
+    },
+    {
+      "epoch": 4.32,
+      "eval_loss": 0.6723945736885071,
+      "eval_runtime": 96.9931,
+      "eval_samples_per_second": 2.062,
+      "eval_steps_per_second": 0.516,
+      "step": 134
+    },
+    {
+      "epoch": 4.352,
+      "grad_norm": 1.607226224281089,
+      "learning_rate": 2e-05,
+      "loss": 0.6168,
+      "step": 135
+    },
+    {
+      "epoch": 4.352,
+      "eval_loss": 0.6719383001327515,
+      "eval_runtime": 97.0819,
+      "eval_samples_per_second": 2.06,
+      "eval_steps_per_second": 0.515,
+      "step": 135
+    },
+    {
+      "epoch": 4.384,
+      "grad_norm": 1.3581330775842726,
+      "learning_rate": 2e-05,
+      "loss": 0.6197,
+      "step": 136
+    },
+    {
+      "epoch": 4.384,
+      "eval_loss": 0.6716668605804443,
+      "eval_runtime": 99.6743,
+      "eval_samples_per_second": 2.007,
+      "eval_steps_per_second": 0.502,
+      "step": 136
+    },
+    {
+      "epoch": 4.416,
+      "grad_norm": 1.6769316310469216,
+      "learning_rate": 2e-05,
+      "loss": 0.638,
+      "step": 137
+    },
+    {
+      "epoch": 4.416,
+      "eval_loss": 0.6708360314369202,
+      "eval_runtime": 95.7987,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 137
+    },
+    {
+      "epoch": 4.448,
+      "grad_norm": 1.5498003031158447,
+      "learning_rate": 2e-05,
+      "loss": 0.6481,
+      "step": 138
+    },
+    {
+      "epoch": 4.448,
+      "eval_loss": 0.6702771782875061,
+      "eval_runtime": 96.0198,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 138
+    },
+    {
+      "epoch": 4.48,
+      "grad_norm": 1.551391635180101,
+      "learning_rate": 2e-05,
+      "loss": 0.6396,
+      "step": 139
+    },
+    {
+      "epoch": 4.48,
+      "eval_loss": 0.6709393262863159,
+      "eval_runtime": 95.7365,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 139
+    },
+    {
+      "epoch": 4.5120000000000005,
+      "grad_norm": 1.534460089474332,
+      "learning_rate": 2e-05,
+      "loss": 0.5738,
+      "step": 140
+    },
+    {
+      "epoch": 4.5120000000000005,
+      "eval_loss": 0.6725634932518005,
+      "eval_runtime": 98.1018,
+      "eval_samples_per_second": 2.039,
+      "eval_steps_per_second": 0.51,
+      "step": 140
+    },
+    {
+      "epoch": 4.5440000000000005,
+      "grad_norm": 1.4820795078494542,
+      "learning_rate": 2e-05,
+      "loss": 0.662,
+      "step": 141
+    },
+    {
+      "epoch": 4.5440000000000005,
+      "eval_loss": 0.6750396490097046,
+      "eval_runtime": 95.8735,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.522,
+      "step": 141
+    },
+    {
+      "epoch": 4.576,
+      "grad_norm": 1.5013493191075218,
+      "learning_rate": 2e-05,
+      "loss": 0.6571,
+      "step": 142
+    },
+    {
+      "epoch": 4.576,
+      "eval_loss": 0.6773877739906311,
+      "eval_runtime": 95.7522,
+      "eval_samples_per_second": 2.089,
+      "eval_steps_per_second": 0.522,
+      "step": 142
+    },
+    {
+      "epoch": 4.608,
+      "grad_norm": 1.5713746383626521,
+      "learning_rate": 2e-05,
+      "loss": 0.6652,
+      "step": 143
+    },
+    {
+      "epoch": 4.608,
+      "eval_loss": 0.677890419960022,
+      "eval_runtime": 95.6097,
+      "eval_samples_per_second": 2.092,
+      "eval_steps_per_second": 0.523,
+      "step": 143
+    },
+    {
+      "epoch": 4.64,
+      "grad_norm": 1.7992319985289944,
+      "learning_rate": 2e-05,
+      "loss": 0.6846,
+      "step": 144
+    },
+    {
+      "epoch": 4.64,
+      "eval_loss": 0.6761859059333801,
+      "eval_runtime": 95.5647,
+      "eval_samples_per_second": 2.093,
+      "eval_steps_per_second": 0.523,
+      "step": 144
+    },
+    {
+      "epoch": 4.672,
+      "grad_norm": 1.8153564044346968,
+      "learning_rate": 2e-05,
+      "loss": 0.6343,
+      "step": 145
+    },
+    {
+      "epoch": 4.672,
+      "eval_loss": 0.6728916764259338,
+      "eval_runtime": 95.7986,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 145
+    },
+    {
+      "epoch": 4.704,
+      "grad_norm": 1.7257180198459412,
+      "learning_rate": 2e-05,
+      "loss": 0.6635,
+      "step": 146
+    },
+    {
+      "epoch": 4.704,
+      "eval_loss": 0.670305073261261,
+      "eval_runtime": 96.0378,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 146
+    },
+    {
+      "epoch": 4.736,
+      "grad_norm": 1.700890150951268,
+      "learning_rate": 2e-05,
+      "loss": 0.6492,
+      "step": 147
+    },
+    {
+      "epoch": 4.736,
+      "eval_loss": 0.6693490743637085,
+      "eval_runtime": 95.7653,
+      "eval_samples_per_second": 2.088,
+      "eval_steps_per_second": 0.522,
+      "step": 147
+    },
+    {
+      "epoch": 4.768,
+      "grad_norm": 1.524592426429662,
+      "learning_rate": 2e-05,
+      "loss": 0.6196,
+      "step": 148
+    },
+    {
+      "epoch": 4.768,
+      "eval_loss": 0.6692916750907898,
+      "eval_runtime": 95.8987,
+      "eval_samples_per_second": 2.086,
+      "eval_steps_per_second": 0.521,
+      "step": 148
+    },
+    {
+      "epoch": 4.8,
+      "grad_norm": 1.765585164658825,
+      "learning_rate": 2e-05,
+      "loss": 0.6634,
+      "step": 149
+    },
+    {
+      "epoch": 4.8,
+      "eval_loss": 0.6684932112693787,
+      "eval_runtime": 96.0068,
+      "eval_samples_per_second": 2.083,
+      "eval_steps_per_second": 0.521,
+      "step": 149
+    },
+    {
+      "epoch": 4.832,
+      "grad_norm": 1.7560172174474262,
+      "learning_rate": 2e-05,
+      "loss": 0.6011,
+      "step": 150
+    },
+    {
+      "epoch": 4.832,
+      "eval_loss": 0.6678715348243713,
+      "eval_runtime": 95.8356,
+      "eval_samples_per_second": 2.087,
+      "eval_steps_per_second": 0.522,
+      "step": 150
+    },
+    {
+      "epoch": 4.864,
+      "grad_norm": 1.5106799279904972,
+      "learning_rate": 2e-05,
+      "loss": 0.6502,
+      "step": 151
+    },
+    {
+      "epoch": 4.864,
+      "eval_loss": 0.6691852807998657,
+      "eval_runtime": 97.5603,
+      "eval_samples_per_second": 2.05,
+      "eval_steps_per_second": 0.513,
+      "step": 151
+    },
+    {
+      "epoch": 4.896,
+      "grad_norm": 1.6515668480995418,
+      "learning_rate": 2e-05,
+      "loss": 0.6207,
+      "step": 152
+    },
+    {
+      "epoch": 4.896,
+      "eval_loss": 0.6714971661567688,
+      "eval_runtime": 96.9472,
+      "eval_samples_per_second": 2.063,
+      "eval_steps_per_second": 0.516,
+      "step": 152
+    },
+    {
+      "epoch": 4.928,
+      "grad_norm": 1.648933444515062,
+      "learning_rate": 2e-05,
+      "loss": 0.6366,
+      "step": 153
+    },
+    {
+      "epoch": 4.928,
+      "eval_loss": 0.6733483672142029,
+      "eval_runtime": 100.8633,
+      "eval_samples_per_second": 1.983,
+      "eval_steps_per_second": 0.496,
+      "step": 153
+    },
+    {
+      "epoch": 4.96,
+      "grad_norm": 1.6749689583250607,
+      "learning_rate": 2e-05,
+      "loss": 0.5815,
+      "step": 154
+    },
+    {
+      "epoch": 4.96,
+      "eval_loss": 0.6746806502342224,
+      "eval_runtime": 100.4948,
+      "eval_samples_per_second": 1.99,
+      "eval_steps_per_second": 0.498,
+      "step": 154
+    },
+    {
+      "epoch": 4.992,
+      "grad_norm": 1.7662303190761401,
+      "learning_rate": 2e-05,
+      "loss": 0.6366,
+      "step": 155
+    },
+    {
+      "epoch": 4.992,
+      "eval_loss": 0.6755189299583435,
+      "eval_runtime": 100.872,
+      "eval_samples_per_second": 1.983,
+      "eval_steps_per_second": 0.496,
+      "step": 155
+    },
+    {
+      "epoch": 4.992,
+      "step": 155,
+      "total_flos": 135390371053568.0,
+      "train_loss": 0.1233276440251258,
+      "train_runtime": 3633.0506,
+      "train_samples_per_second": 1.376,
+      "train_steps_per_second": 0.043
+    }
+  ],
+  "logging_steps": 1.0,
+  "max_steps": 155,
+  "num_input_tokens_seen": 0,
+  "num_train_epochs": 5,
+  "save_steps": 5,
+  "stateful_callbacks": {
+    "TrainerControl": {
+      "args": {
+        "should_epoch_stop": false,
+        "should_evaluate": false,
+        "should_log": false,
+        "should_save": true,
+        "should_training_stop": true
+      },
+      "attributes": {}
+    }
+  },
+  "total_flos": 135390371053568.0,
+  "train_batch_size": 4,
+  "trial_name": null,
+  "trial_params": null
+}