just1nseo commited on
Commit
cf85808
1 Parent(s): 12dcbdb

Model save

Browse files
README.md ADDED
@@ -0,0 +1,89 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: alignment-handbook/zephyr-7b-sft-full
3
+ library_name: peft
4
+ license: apache-2.0
5
+ tags:
6
+ - trl
7
+ - dpo
8
+ - generated_from_trainer
9
+ model-index:
10
+ - name: zephyr-dpop-qlora-uf-ours-5e-7
11
+ results: []
12
+ ---
13
+
14
+ <!-- This model card has been generated automatically according to the information the Trainer had access to. You
15
+ should probably proofread and complete it, then remove this comment. -->
16
+
17
+ # zephyr-dpop-qlora-uf-ours-5e-7
18
+
19
+ This model is a fine-tuned version of [alignment-handbook/zephyr-7b-sft-full](https://huggingface.co/alignment-handbook/zephyr-7b-sft-full) on the None dataset.
20
+ It achieves the following results on the evaluation set:
21
+ - Loss: 0.9543
22
+ - Positive Losses: 2.5810
23
+ - Dpo Losses: 0.6658
24
+ - Rewards/chosen: 0.0598
25
+ - Rewards/rejected: -0.0042
26
+ - Rewards/accuracies: 0.6290
27
+ - Rewards/margins: 0.0640
28
+ - Rewards/margins Max: 0.3475
29
+ - Rewards/margins Min: -0.1826
30
+ - Rewards/margins Std: 0.1767
31
+ - Logps/rejected: -259.0028
32
+ - Logps/chosen: -278.6134
33
+ - Logits/rejected: -2.6749
34
+ - Logits/chosen: -2.7127
35
+
36
+ ## Model description
37
+
38
+ More information needed
39
+
40
+ ## Intended uses & limitations
41
+
42
+ More information needed
43
+
44
+ ## Training and evaluation data
45
+
46
+ More information needed
47
+
48
+ ## Training procedure
49
+
50
+ ### Training hyperparameters
51
+
52
+ The following hyperparameters were used during training:
53
+ - learning_rate: 5e-07
54
+ - train_batch_size: 4
55
+ - eval_batch_size: 8
56
+ - seed: 42
57
+ - distributed_type: multi-GPU
58
+ - num_devices: 2
59
+ - gradient_accumulation_steps: 2
60
+ - total_train_batch_size: 16
61
+ - total_eval_batch_size: 16
62
+ - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
63
+ - lr_scheduler_type: cosine
64
+ - lr_scheduler_warmup_ratio: 0.1
65
+ - num_epochs: 3
66
+
67
+ ### Training results
68
+
69
+ | Training Loss | Epoch | Step | Validation Loss | Positive Losses | Dpo Losses | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Rewards/margins Max | Rewards/margins Min | Rewards/margins Std | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
70
+ |:-------------:|:-----:|:----:|:---------------:|:---------------:|:----------:|:--------------:|:----------------:|:------------------:|:---------------:|:-------------------:|:-------------------:|:-------------------:|:--------------:|:------------:|:---------------:|:-------------:|
71
+ | 0.6884 | 0.28 | 100 | 0.6931 | 0.0111 | 0.6918 | 0.0136 | 0.0108 | 0.6080 | 0.0028 | 0.0179 | -0.0107 | 0.0094 | -257.4949 | -283.2318 | -2.7651 | -2.8042 |
72
+ | 0.6627 | 0.56 | 200 | 0.6995 | 0.1223 | 0.6858 | 0.0465 | 0.0311 | 0.5960 | 0.0153 | 0.0899 | -0.0496 | 0.0465 | -255.4640 | -279.9481 | -2.7485 | -2.7871 |
73
+ | 0.6293 | 0.85 | 300 | 0.7193 | 0.3552 | 0.6803 | 0.0675 | 0.0398 | 0.5960 | 0.0278 | 0.1601 | -0.0863 | 0.0826 | -254.6033 | -277.8385 | -2.7306 | -2.7684 |
74
+ | 0.6236 | 1.13 | 400 | 0.7519 | 0.6894 | 0.6756 | 0.0800 | 0.0412 | 0.6090 | 0.0388 | 0.2182 | -0.1140 | 0.1113 | -254.4585 | -276.5968 | -2.7119 | -2.7494 |
75
+ | 0.6009 | 1.41 | 500 | 0.8434 | 1.5495 | 0.6718 | 0.0639 | 0.0154 | 0.6090 | 0.0484 | 0.2709 | -0.1440 | 0.1389 | -257.0343 | -278.2061 | -2.6920 | -2.7295 |
76
+ | 0.6136 | 1.69 | 600 | 0.8727 | 1.8302 | 0.6691 | 0.0687 | 0.0134 | 0.6130 | 0.0553 | 0.3049 | -0.1595 | 0.1553 | -257.2360 | -277.7244 | -2.6827 | -2.7203 |
77
+ | 0.5918 | 1.97 | 700 | 0.8998 | 2.0811 | 0.6677 | 0.0671 | 0.0081 | 0.6220 | 0.0591 | 0.3231 | -0.1685 | 0.1641 | -257.7734 | -277.8808 | -2.6797 | -2.7172 |
78
+ | 0.5636 | 2.25 | 800 | 0.9371 | 2.4201 | 0.6667 | 0.0611 | -0.0007 | 0.6260 | 0.0618 | 0.3370 | -0.1777 | 0.1716 | -258.6473 | -278.4820 | -2.6734 | -2.7116 |
79
+ | 0.5736 | 2.54 | 900 | 0.9591 | 2.6268 | 0.6659 | 0.0578 | -0.0060 | 0.6320 | 0.0639 | 0.3467 | -0.1823 | 0.1764 | -259.1817 | -278.8090 | -2.6726 | -2.7107 |
80
+ | 0.5825 | 2.82 | 1000 | 0.9543 | 2.5810 | 0.6658 | 0.0598 | -0.0042 | 0.6290 | 0.0640 | 0.3475 | -0.1826 | 0.1767 | -259.0028 | -278.6134 | -2.6749 | -2.7127 |
81
+
82
+
83
+ ### Framework versions
84
+
85
+ - PEFT 0.7.1
86
+ - Transformers 4.39.0.dev0
87
+ - Pytorch 2.1.2+cu121
88
+ - Datasets 2.14.6
89
+ - Tokenizers 0.15.2
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:415354ee65160701ef2e28aecf31ef3c64e321f9c19ffe2d066298e28de3ed1f
3
  size 671150064
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a502f0a27d352c5c0b247cc472d2603fe33e99f392870efa94152007030602bc
3
  size 671150064
all_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 3.0,
3
+ "train_loss": 0.6146632369135467,
4
+ "train_runtime": 13546.83,
5
+ "train_samples": 5678,
6
+ "train_samples_per_second": 1.257,
7
+ "train_steps_per_second": 0.079
8
+ }
runs/Jul15_16-54-40_notebook-deployment-48-7d9b6c99-p5kv4/events.out.tfevents.1721062520.notebook-deployment-48-7d9b6c99-p5kv4.52454.0 CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:014455845f46996194d58d64450c341a1207d817815b38ac7b041a6155532768
3
- size 116029
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1ff30c43e44e6b01f35141c068d816669e4d4632dbb42180378b807eecdd06d7
3
+ size 122353
train_results.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "epoch": 3.0,
3
+ "train_loss": 0.6146632369135467,
4
+ "train_runtime": 13546.83,
5
+ "train_samples": 5678,
6
+ "train_samples_per_second": 1.257,
7
+ "train_steps_per_second": 0.079
8
+ }
trainer_state.json ADDED
@@ -0,0 +1,2380 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "best_metric": null,
3
+ "best_model_checkpoint": null,
4
+ "epoch": 3.0,
5
+ "eval_steps": 100,
6
+ "global_step": 1065,
7
+ "is_hyper_param_search": false,
8
+ "is_local_process_zero": true,
9
+ "is_world_process_zero": true,
10
+ "log_history": [
11
+ {
12
+ "dpo_losses": 0.6931471824645996,
13
+ "epoch": 0.0,
14
+ "grad_norm": 1.601693496929256,
15
+ "learning_rate": 4.6728971962616815e-09,
16
+ "logits/chosen": -2.861618995666504,
17
+ "logits/rejected": -2.8205904960632324,
18
+ "logps/chosen": -271.06011962890625,
19
+ "logps/rejected": -211.1704559326172,
20
+ "loss": 0.6931,
21
+ "positive_losses": 0.0,
22
+ "rewards/accuracies": 0.0,
23
+ "rewards/chosen": 0.0,
24
+ "rewards/margins": 0.0,
25
+ "rewards/margins_max": 0.0,
26
+ "rewards/margins_min": 0.0,
27
+ "rewards/margins_std": 0.0,
28
+ "rewards/rejected": 0.0,
29
+ "step": 1
30
+ },
31
+ {
32
+ "dpo_losses": 0.6933066844940186,
33
+ "epoch": 0.03,
34
+ "grad_norm": 20.410276988279243,
35
+ "learning_rate": 4.672897196261682e-08,
36
+ "logits/chosen": -2.8335423469543457,
37
+ "logits/rejected": -2.7909910678863525,
38
+ "logps/chosen": -325.0599365234375,
39
+ "logps/rejected": -274.9435119628906,
40
+ "loss": 0.6965,
41
+ "positive_losses": 0.050453320145606995,
42
+ "rewards/accuracies": 0.2638888955116272,
43
+ "rewards/chosen": -9.644959936849773e-05,
44
+ "rewards/margins": -0.00031749275512993336,
45
+ "rewards/margins_max": 0.0014122920110821724,
46
+ "rewards/margins_min": -0.0021905910689383745,
47
+ "rewards/margins_std": 0.0016328593483194709,
48
+ "rewards/rejected": 0.00022104315576143563,
49
+ "step": 10
50
+ },
51
+ {
52
+ "dpo_losses": 0.6930336952209473,
53
+ "epoch": 0.06,
54
+ "grad_norm": 21.230310271372627,
55
+ "learning_rate": 9.345794392523364e-08,
56
+ "logits/chosen": -2.725196361541748,
57
+ "logits/rejected": -2.706851005554199,
58
+ "logps/chosen": -293.7933654785156,
59
+ "logps/rejected": -215.7693328857422,
60
+ "loss": 0.6994,
61
+ "positive_losses": 0.07486093044281006,
62
+ "rewards/accuracies": 0.5375000238418579,
63
+ "rewards/chosen": 0.00032553955679759383,
64
+ "rewards/margins": 0.00022889305546414107,
65
+ "rewards/margins_max": 0.002858394058421254,
66
+ "rewards/margins_min": -0.0026940270327031612,
67
+ "rewards/margins_std": 0.0024744768161326647,
68
+ "rewards/rejected": 9.664653771324083e-05,
69
+ "step": 20
70
+ },
71
+ {
72
+ "dpo_losses": 0.6927663087844849,
73
+ "epoch": 0.08,
74
+ "grad_norm": 17.85805321688973,
75
+ "learning_rate": 1.4018691588785045e-07,
76
+ "logits/chosen": -2.819422960281372,
77
+ "logits/rejected": -2.750398874282837,
78
+ "logps/chosen": -303.5675048828125,
79
+ "logps/rejected": -232.4523162841797,
80
+ "loss": 0.6975,
81
+ "positive_losses": 0.04014534875750542,
82
+ "rewards/accuracies": 0.612500011920929,
83
+ "rewards/chosen": 0.0013837231090292335,
84
+ "rewards/margins": 0.0007641493575647473,
85
+ "rewards/margins_max": 0.003605480073019862,
86
+ "rewards/margins_min": -0.0024470784701406956,
87
+ "rewards/margins_std": 0.002705852035433054,
88
+ "rewards/rejected": 0.000619573867879808,
89
+ "step": 30
90
+ },
91
+ {
92
+ "dpo_losses": 0.6928492188453674,
93
+ "epoch": 0.11,
94
+ "grad_norm": 1.7332928692545067,
95
+ "learning_rate": 1.8691588785046729e-07,
96
+ "logits/chosen": -2.8429579734802246,
97
+ "logits/rejected": -2.7627620697021484,
98
+ "logps/chosen": -277.1645202636719,
99
+ "logps/rejected": -225.4744415283203,
100
+ "loss": 0.6937,
101
+ "positive_losses": 0.007175350096076727,
102
+ "rewards/accuracies": 0.5625,
103
+ "rewards/chosen": 0.003201347542926669,
104
+ "rewards/margins": 0.0005985596217215061,
105
+ "rewards/margins_max": 0.003463293891400099,
106
+ "rewards/margins_min": -0.00237638084217906,
107
+ "rewards/margins_std": 0.0026152683421969414,
108
+ "rewards/rejected": 0.0026027881540358067,
109
+ "step": 40
110
+ },
111
+ {
112
+ "dpo_losses": 0.6924293637275696,
113
+ "epoch": 0.14,
114
+ "grad_norm": 2.2231645321380933,
115
+ "learning_rate": 2.336448598130841e-07,
116
+ "logits/chosen": -2.8112452030181885,
117
+ "logits/rejected": -2.7390694618225098,
118
+ "logps/chosen": -275.0422058105469,
119
+ "logps/rejected": -233.49331665039062,
120
+ "loss": 0.6929,
121
+ "positive_losses": 0.007281684782356024,
122
+ "rewards/accuracies": 0.6625000238418579,
123
+ "rewards/chosen": 0.0052792662754654884,
124
+ "rewards/margins": 0.001439323415979743,
125
+ "rewards/margins_max": 0.0054040043614804745,
126
+ "rewards/margins_min": -0.0021666125394403934,
127
+ "rewards/margins_std": 0.003360858652740717,
128
+ "rewards/rejected": 0.003839943092316389,
129
+ "step": 50
130
+ },
131
+ {
132
+ "dpo_losses": 0.6919517517089844,
133
+ "epoch": 0.17,
134
+ "grad_norm": 2.413825162312939,
135
+ "learning_rate": 2.803738317757009e-07,
136
+ "logits/chosen": -2.7509729862213135,
137
+ "logits/rejected": -2.7115964889526367,
138
+ "logps/chosen": -312.9291076660156,
139
+ "logps/rejected": -262.30059814453125,
140
+ "loss": 0.6935,
141
+ "positive_losses": 0.00518798828125,
142
+ "rewards/accuracies": 0.699999988079071,
143
+ "rewards/chosen": 0.007105088327080011,
144
+ "rewards/margins": 0.002395912306383252,
145
+ "rewards/margins_max": 0.00661232927814126,
146
+ "rewards/margins_min": -0.000950089015532285,
147
+ "rewards/margins_std": 0.0033020805567502975,
148
+ "rewards/rejected": 0.004709175787866116,
149
+ "step": 60
150
+ },
151
+ {
152
+ "dpo_losses": 0.6914928555488586,
153
+ "epoch": 0.2,
154
+ "grad_norm": 1.8246267381191894,
155
+ "learning_rate": 3.271028037383177e-07,
156
+ "logits/chosen": -2.78471302986145,
157
+ "logits/rejected": -2.722522258758545,
158
+ "logps/chosen": -321.91546630859375,
159
+ "logps/rejected": -235.0277557373047,
160
+ "loss": 0.6913,
161
+ "positive_losses": 0.0,
162
+ "rewards/accuracies": 0.7124999761581421,
163
+ "rewards/chosen": 0.009457603096961975,
164
+ "rewards/margins": 0.0033192108385264874,
165
+ "rewards/margins_max": 0.009357670322060585,
166
+ "rewards/margins_min": -0.001532487804070115,
167
+ "rewards/margins_std": 0.0049352385103702545,
168
+ "rewards/rejected": 0.006138390861451626,
169
+ "step": 70
170
+ },
171
+ {
172
+ "dpo_losses": 0.6906715631484985,
173
+ "epoch": 0.23,
174
+ "grad_norm": 1.8774986784177312,
175
+ "learning_rate": 3.7383177570093457e-07,
176
+ "logits/chosen": -2.8345515727996826,
177
+ "logits/rejected": -2.7776432037353516,
178
+ "logps/chosen": -320.06634521484375,
179
+ "logps/rejected": -285.67156982421875,
180
+ "loss": 0.6909,
181
+ "positive_losses": 0.0006896972772665322,
182
+ "rewards/accuracies": 0.737500011920929,
183
+ "rewards/chosen": 0.012041566893458366,
184
+ "rewards/margins": 0.004970206413418055,
185
+ "rewards/margins_max": 0.013391142711043358,
186
+ "rewards/margins_min": -0.0014745177468284965,
187
+ "rewards/margins_std": 0.006737919058650732,
188
+ "rewards/rejected": 0.007071360945701599,
189
+ "step": 80
190
+ },
191
+ {
192
+ "dpo_losses": 0.6892363429069519,
193
+ "epoch": 0.25,
194
+ "grad_norm": 2.6194297181523774,
195
+ "learning_rate": 4.205607476635514e-07,
196
+ "logits/chosen": -2.7694807052612305,
197
+ "logits/rejected": -2.726945400238037,
198
+ "logps/chosen": -274.4305725097656,
199
+ "logps/rejected": -198.20901489257812,
200
+ "loss": 0.6894,
201
+ "positive_losses": 0.0005071639898233116,
202
+ "rewards/accuracies": 0.8125,
203
+ "rewards/chosen": 0.015641968697309494,
204
+ "rewards/margins": 0.007860125042498112,
205
+ "rewards/margins_max": 0.01880500093102455,
206
+ "rewards/margins_min": -0.0005177496350370347,
207
+ "rewards/margins_std": 0.008911145851016045,
208
+ "rewards/rejected": 0.007781843654811382,
209
+ "step": 90
210
+ },
211
+ {
212
+ "dpo_losses": 0.6890109777450562,
213
+ "epoch": 0.28,
214
+ "grad_norm": 1.9205518680432034,
215
+ "learning_rate": 4.672897196261682e-07,
216
+ "logits/chosen": -2.80806303024292,
217
+ "logits/rejected": -2.7697219848632812,
218
+ "logps/chosen": -279.78045654296875,
219
+ "logps/rejected": -261.3872375488281,
220
+ "loss": 0.6884,
221
+ "positive_losses": 0.0013298034900799394,
222
+ "rewards/accuracies": 0.7875000238418579,
223
+ "rewards/chosen": 0.016312729567289352,
224
+ "rewards/margins": 0.00831080786883831,
225
+ "rewards/margins_max": 0.018694214522838593,
226
+ "rewards/margins_min": 0.00036361132515594363,
227
+ "rewards/margins_std": 0.00843831431120634,
228
+ "rewards/rejected": 0.008001920767128468,
229
+ "step": 100
230
+ },
231
+ {
232
+ "epoch": 0.28,
233
+ "eval_dpo_losses": 0.6917725205421448,
234
+ "eval_logits/chosen": -2.8041651248931885,
235
+ "eval_logits/rejected": -2.7650842666625977,
236
+ "eval_logps/chosen": -283.2318420410156,
237
+ "eval_logps/rejected": -257.4949035644531,
238
+ "eval_loss": 0.6931358575820923,
239
+ "eval_positive_losses": 0.011085770092904568,
240
+ "eval_rewards/accuracies": 0.6079999804496765,
241
+ "eval_rewards/chosen": 0.0136156240478158,
242
+ "eval_rewards/margins": 0.0027753659524023533,
243
+ "eval_rewards/margins_max": 0.017850197851657867,
244
+ "eval_rewards/margins_min": -0.010674619115889072,
245
+ "eval_rewards/margins_std": 0.009409897960722446,
246
+ "eval_rewards/rejected": 0.010840258561074734,
247
+ "eval_runtime": 428.0491,
248
+ "eval_samples_per_second": 4.672,
249
+ "eval_steps_per_second": 0.292,
250
+ "step": 100
251
+ },
252
+ {
253
+ "dpo_losses": 0.6845903396606445,
254
+ "epoch": 0.31,
255
+ "grad_norm": 2.377556980384867,
256
+ "learning_rate": 4.999879018839287e-07,
257
+ "logits/chosen": -2.8750226497650146,
258
+ "logits/rejected": -2.7789218425750732,
259
+ "logps/chosen": -372.0056457519531,
260
+ "logps/rejected": -285.5588073730469,
261
+ "loss": 0.6848,
262
+ "positive_losses": 0.0,
263
+ "rewards/accuracies": 0.9125000238418579,
264
+ "rewards/chosen": 0.026000410318374634,
265
+ "rewards/margins": 0.017250509932637215,
266
+ "rewards/margins_max": 0.0338827520608902,
267
+ "rewards/margins_min": 0.0025702244602143764,
268
+ "rewards/margins_std": 0.01393597386777401,
269
+ "rewards/rejected": 0.008749897591769695,
270
+ "step": 110
271
+ },
272
+ {
273
+ "dpo_losses": 0.6825487613677979,
274
+ "epoch": 0.34,
275
+ "grad_norm": 2.115970951796753,
276
+ "learning_rate": 4.997728568369408e-07,
277
+ "logits/chosen": -2.844456911087036,
278
+ "logits/rejected": -2.790116786956787,
279
+ "logps/chosen": -345.28887939453125,
280
+ "logps/rejected": -265.2372741699219,
281
+ "loss": 0.6828,
282
+ "positive_losses": 0.0,
283
+ "rewards/accuracies": 0.8999999761581421,
284
+ "rewards/chosen": 0.03097311221063137,
285
+ "rewards/margins": 0.02140369825065136,
286
+ "rewards/margins_max": 0.042990267276763916,
287
+ "rewards/margins_min": 0.005240675527602434,
288
+ "rewards/margins_std": 0.017378441989421844,
289
+ "rewards/rejected": 0.009569412097334862,
290
+ "step": 120
291
+ },
292
+ {
293
+ "dpo_losses": 0.679545521736145,
294
+ "epoch": 0.37,
295
+ "grad_norm": 1.8392631764823357,
296
+ "learning_rate": 4.992892309373227e-07,
297
+ "logits/chosen": -2.8388783931732178,
298
+ "logits/rejected": -2.7566184997558594,
299
+ "logps/chosen": -378.5120849609375,
300
+ "logps/rejected": -299.33154296875,
301
+ "loss": 0.6796,
302
+ "positive_losses": 0.00618324289098382,
303
+ "rewards/accuracies": 0.9125000238418579,
304
+ "rewards/chosen": 0.0385219044983387,
305
+ "rewards/margins": 0.027528375387191772,
306
+ "rewards/margins_max": 0.05331053584814072,
307
+ "rewards/margins_min": 0.003396461484953761,
308
+ "rewards/margins_std": 0.021965984255075455,
309
+ "rewards/rejected": 0.01099353563040495,
310
+ "step": 130
311
+ },
312
+ {
313
+ "dpo_losses": 0.6778298020362854,
314
+ "epoch": 0.39,
315
+ "grad_norm": 1.4924163642849055,
316
+ "learning_rate": 4.985375442281968e-07,
317
+ "logits/chosen": -2.791003704071045,
318
+ "logits/rejected": -2.751797914505005,
319
+ "logps/chosen": -314.1080322265625,
320
+ "logps/rejected": -268.6662902832031,
321
+ "loss": 0.6798,
322
+ "positive_losses": 0.0,
323
+ "rewards/accuracies": 0.9125000238418579,
324
+ "rewards/chosen": 0.04306010901927948,
325
+ "rewards/margins": 0.03122769668698311,
326
+ "rewards/margins_max": 0.06796383112668991,
327
+ "rewards/margins_min": 0.004374162759631872,
328
+ "rewards/margins_std": 0.02954345941543579,
329
+ "rewards/rejected": 0.011832410469651222,
330
+ "step": 140
331
+ },
332
+ {
333
+ "dpo_losses": 0.6768456101417542,
334
+ "epoch": 0.42,
335
+ "grad_norm": 2.003305207476629,
336
+ "learning_rate": 4.975186049985817e-07,
337
+ "logits/chosen": -2.7922985553741455,
338
+ "logits/rejected": -2.752359628677368,
339
+ "logps/chosen": -284.1427307128906,
340
+ "logps/rejected": -211.85299682617188,
341
+ "loss": 0.6759,
342
+ "positive_losses": 0.000370025634765625,
343
+ "rewards/accuracies": 0.9125000238418579,
344
+ "rewards/chosen": 0.04542272537946701,
345
+ "rewards/margins": 0.033125050365924835,
346
+ "rewards/margins_max": 0.06794625520706177,
347
+ "rewards/margins_min": 0.004188986029475927,
348
+ "rewards/margins_std": 0.029155880212783813,
349
+ "rewards/rejected": 0.012297680601477623,
350
+ "step": 150
351
+ },
352
+ {
353
+ "dpo_losses": 0.6737908720970154,
354
+ "epoch": 0.45,
355
+ "grad_norm": 1.673622074152846,
356
+ "learning_rate": 4.962335089142375e-07,
357
+ "logits/chosen": -2.770982265472412,
358
+ "logits/rejected": -2.716006278991699,
359
+ "logps/chosen": -295.1300048828125,
360
+ "logps/rejected": -232.91378784179688,
361
+ "loss": 0.6732,
362
+ "positive_losses": 0.00577545166015625,
363
+ "rewards/accuracies": 0.9375,
364
+ "rewards/chosen": 0.054168350994586945,
365
+ "rewards/margins": 0.03937327116727829,
366
+ "rewards/margins_max": 0.0770280510187149,
367
+ "rewards/margins_min": 0.005630848463624716,
368
+ "rewards/margins_std": 0.03246045857667923,
369
+ "rewards/rejected": 0.014795074239373207,
370
+ "step": 160
371
+ },
372
+ {
373
+ "dpo_losses": 0.6671137809753418,
374
+ "epoch": 0.48,
375
+ "grad_norm": 1.849316679565777,
376
+ "learning_rate": 4.946836378394966e-07,
377
+ "logits/chosen": -2.8234620094299316,
378
+ "logits/rejected": -2.76485276222229,
379
+ "logps/chosen": -323.1846923828125,
380
+ "logps/rejected": -295.15374755859375,
381
+ "loss": 0.6687,
382
+ "positive_losses": 0.0,
383
+ "rewards/accuracies": 0.887499988079071,
384
+ "rewards/chosen": 0.06436900794506073,
385
+ "rewards/margins": 0.05324960872530937,
386
+ "rewards/margins_max": 0.1005321741104126,
387
+ "rewards/margins_min": 0.010768366977572441,
388
+ "rewards/margins_std": 0.04033435881137848,
389
+ "rewards/rejected": 0.011119391769170761,
390
+ "step": 170
391
+ },
392
+ {
393
+ "dpo_losses": 0.6649172902107239,
394
+ "epoch": 0.51,
395
+ "grad_norm": 1.9724747678160601,
396
+ "learning_rate": 4.92870658351344e-07,
397
+ "logits/chosen": -2.815851926803589,
398
+ "logits/rejected": -2.727149248123169,
399
+ "logps/chosen": -324.7388916015625,
400
+ "logps/rejected": -292.8143615722656,
401
+ "loss": 0.6646,
402
+ "positive_losses": 0.0,
403
+ "rewards/accuracies": 0.9375,
404
+ "rewards/chosen": 0.07161912322044373,
405
+ "rewards/margins": 0.057913679629564285,
406
+ "rewards/margins_max": 0.1177336573600769,
407
+ "rewards/margins_min": 0.01562586799263954,
408
+ "rewards/margins_std": 0.04572517052292824,
409
+ "rewards/rejected": 0.013705444522202015,
410
+ "step": 180
411
+ },
412
+ {
413
+ "dpo_losses": 0.6602508425712585,
414
+ "epoch": 0.54,
415
+ "grad_norm": 2.366543924703805,
416
+ "learning_rate": 4.90796519947347e-07,
417
+ "logits/chosen": -2.776212215423584,
418
+ "logits/rejected": -2.6835803985595703,
419
+ "logps/chosen": -313.767578125,
420
+ "logps/rejected": -236.1117706298828,
421
+ "loss": 0.6621,
422
+ "positive_losses": 0.008851242251694202,
423
+ "rewards/accuracies": 0.9125000238418579,
424
+ "rewards/chosen": 0.08193298429250717,
425
+ "rewards/margins": 0.0681738331913948,
426
+ "rewards/margins_max": 0.1408631056547165,
427
+ "rewards/margins_min": 0.011917511001229286,
428
+ "rewards/margins_std": 0.059797000139951706,
429
+ "rewards/rejected": 0.013759145513176918,
430
+ "step": 190
431
+ },
432
+ {
433
+ "dpo_losses": 0.659784197807312,
434
+ "epoch": 0.56,
435
+ "grad_norm": 8.125986362153837,
436
+ "learning_rate": 4.88463452949359e-07,
437
+ "logits/chosen": -2.851494312286377,
438
+ "logits/rejected": -2.774346113204956,
439
+ "logps/chosen": -330.9751281738281,
440
+ "logps/rejected": -263.9167175292969,
441
+ "loss": 0.6627,
442
+ "positive_losses": 0.008233070373535156,
443
+ "rewards/accuracies": 0.925000011920929,
444
+ "rewards/chosen": 0.08557725697755814,
445
+ "rewards/margins": 0.06877057254314423,
446
+ "rewards/margins_max": 0.13887056708335876,
447
+ "rewards/margins_min": 0.01435057632625103,
448
+ "rewards/margins_std": 0.05443096160888672,
449
+ "rewards/rejected": 0.016806693747639656,
450
+ "step": 200
451
+ },
452
+ {
453
+ "epoch": 0.56,
454
+ "eval_dpo_losses": 0.6858174800872803,
455
+ "eval_logits/chosen": -2.7871391773223877,
456
+ "eval_logits/rejected": -2.7485153675079346,
457
+ "eval_logps/chosen": -279.9480895996094,
458
+ "eval_logps/rejected": -255.46397399902344,
459
+ "eval_loss": 0.6995190382003784,
460
+ "eval_positive_losses": 0.1222764253616333,
461
+ "eval_rewards/accuracies": 0.5960000157356262,
462
+ "eval_rewards/chosen": 0.04645317792892456,
463
+ "eval_rewards/margins": 0.015303360298275948,
464
+ "eval_rewards/margins_max": 0.08988457173109055,
465
+ "eval_rewards/margins_min": -0.049583785235881805,
466
+ "eval_rewards/margins_std": 0.04650707170367241,
467
+ "eval_rewards/rejected": 0.031149819493293762,
468
+ "eval_runtime": 428.2785,
469
+ "eval_samples_per_second": 4.67,
470
+ "eval_steps_per_second": 0.292,
471
+ "step": 200
472
+ },
473
+ {
474
+ "dpo_losses": 0.6529717445373535,
475
+ "epoch": 0.59,
476
+ "grad_norm": 1.9509652365320336,
477
+ "learning_rate": 4.858739661052539e-07,
478
+ "logits/chosen": -2.7506725788116455,
479
+ "logits/rejected": -2.722893238067627,
480
+ "logps/chosen": -310.4784240722656,
481
+ "logps/rejected": -255.10678100585938,
482
+ "loss": 0.6562,
483
+ "positive_losses": 0.017461013048887253,
484
+ "rewards/accuracies": 0.925000011920929,
485
+ "rewards/chosen": 0.09705875813961029,
486
+ "rewards/margins": 0.08387880027294159,
487
+ "rewards/margins_max": 0.17185379564762115,
488
+ "rewards/margins_min": 0.010865801945328712,
489
+ "rewards/margins_std": 0.07355490326881409,
490
+ "rewards/rejected": 0.013179955072700977,
491
+ "step": 210
492
+ },
493
+ {
494
+ "dpo_losses": 0.6484541893005371,
495
+ "epoch": 0.62,
496
+ "grad_norm": 1.8456619177361002,
497
+ "learning_rate": 4.830308438912687e-07,
498
+ "logits/chosen": -2.8496181964874268,
499
+ "logits/rejected": -2.7542800903320312,
500
+ "logps/chosen": -342.6438293457031,
501
+ "logps/rejected": -272.774169921875,
502
+ "loss": 0.6524,
503
+ "positive_losses": 0.0,
504
+ "rewards/accuracies": 0.949999988079071,
505
+ "rewards/chosen": 0.10812418162822723,
506
+ "rewards/margins": 0.0925908237695694,
507
+ "rewards/margins_max": 0.1654980480670929,
508
+ "rewards/margins_min": 0.022400889545679092,
509
+ "rewards/margins_std": 0.06456473469734192,
510
+ "rewards/rejected": 0.015533369965851307,
511
+ "step": 220
512
+ },
513
+ {
514
+ "dpo_losses": 0.6464593410491943,
515
+ "epoch": 0.65,
516
+ "grad_norm": 1.9535872255784013,
517
+ "learning_rate": 4.799371435178545e-07,
518
+ "logits/chosen": -2.849377393722534,
519
+ "logits/rejected": -2.7559587955474854,
520
+ "logps/chosen": -360.3794860839844,
521
+ "logps/rejected": -260.52764892578125,
522
+ "loss": 0.6444,
523
+ "positive_losses": 0.015559768304228783,
524
+ "rewards/accuracies": 0.9375,
525
+ "rewards/chosen": 0.1129622682929039,
526
+ "rewards/margins": 0.09730460494756699,
527
+ "rewards/margins_max": 0.18746943771839142,
528
+ "rewards/margins_min": 0.02147417888045311,
529
+ "rewards/margins_std": 0.07491712272167206,
530
+ "rewards/rejected": 0.015657661482691765,
531
+ "step": 230
532
+ },
533
+ {
534
+ "dpo_losses": 0.653369128704071,
535
+ "epoch": 0.68,
536
+ "grad_norm": 1.576015913761948,
537
+ "learning_rate": 4.765961916422574e-07,
538
+ "logits/chosen": -2.7950048446655273,
539
+ "logits/rejected": -2.725090742111206,
540
+ "logps/chosen": -314.9557800292969,
541
+ "logps/rejected": -281.58880615234375,
542
+ "loss": 0.6537,
543
+ "positive_losses": 0.016252517700195312,
544
+ "rewards/accuracies": 0.875,
545
+ "rewards/chosen": 0.09841253608465195,
546
+ "rewards/margins": 0.08255226910114288,
547
+ "rewards/margins_max": 0.16673865914344788,
548
+ "rewards/margins_min": 0.015284004621207714,
549
+ "rewards/margins_std": 0.06914026290178299,
550
+ "rewards/rejected": 0.015860268846154213,
551
+ "step": 240
552
+ },
553
+ {
554
+ "dpo_losses": 0.6478245258331299,
555
+ "epoch": 0.7,
556
+ "grad_norm": 1.8968881320655915,
557
+ "learning_rate": 4.730115807913626e-07,
558
+ "logits/chosen": -2.792325258255005,
559
+ "logits/rejected": -2.715862989425659,
560
+ "logps/chosen": -300.74859619140625,
561
+ "logps/rejected": -249.9875030517578,
562
+ "loss": 0.6546,
563
+ "positive_losses": 0.023810099810361862,
564
+ "rewards/accuracies": 0.9375,
565
+ "rewards/chosen": 0.1053255945444107,
566
+ "rewards/margins": 0.09475791454315186,
567
+ "rewards/margins_max": 0.19134044647216797,
568
+ "rewards/margins_min": 0.01620624028146267,
569
+ "rewards/margins_std": 0.08089859038591385,
570
+ "rewards/rejected": 0.01056766789406538,
571
+ "step": 250
572
+ },
573
+ {
574
+ "dpo_losses": 0.6441227197647095,
575
+ "epoch": 0.73,
576
+ "grad_norm": 1.7119013888667354,
577
+ "learning_rate": 4.691871654986485e-07,
578
+ "logits/chosen": -2.839354991912842,
579
+ "logits/rejected": -2.756789445877075,
580
+ "logps/chosen": -280.3956298828125,
581
+ "logps/rejected": -257.5377197265625,
582
+ "loss": 0.6395,
583
+ "positive_losses": 0.004910898394882679,
584
+ "rewards/accuracies": 0.949999988079071,
585
+ "rewards/chosen": 0.11697886139154434,
586
+ "rewards/margins": 0.10304615646600723,
587
+ "rewards/margins_max": 0.20939341187477112,
588
+ "rewards/margins_min": 0.02202121913433075,
589
+ "rewards/margins_std": 0.08740874379873276,
590
+ "rewards/rejected": 0.01393270492553711,
591
+ "step": 260
592
+ },
593
+ {
594
+ "dpo_losses": 0.6456023454666138,
595
+ "epoch": 0.76,
596
+ "grad_norm": 1.984892770095366,
597
+ "learning_rate": 4.6512705815940536e-07,
598
+ "logits/chosen": -2.783090591430664,
599
+ "logits/rejected": -2.730664014816284,
600
+ "logps/chosen": -297.0513000488281,
601
+ "logps/rejected": -260.38958740234375,
602
+ "loss": 0.6522,
603
+ "positive_losses": 0.08494539558887482,
604
+ "rewards/accuracies": 0.9125000238418579,
605
+ "rewards/chosen": 0.12064293771982193,
606
+ "rewards/margins": 0.09954321384429932,
607
+ "rewards/margins_max": 0.18979784846305847,
608
+ "rewards/margins_min": 0.028997600078582764,
609
+ "rewards/margins_std": 0.07252895832061768,
610
+ "rewards/rejected": 0.021099697798490524,
611
+ "step": 270
612
+ },
613
+ {
614
+ "dpo_losses": 0.6418065428733826,
615
+ "epoch": 0.79,
616
+ "grad_norm": 1.7581310746474135,
617
+ "learning_rate": 4.6083562460867544e-07,
618
+ "logits/chosen": -2.75692081451416,
619
+ "logits/rejected": -2.717991590499878,
620
+ "logps/chosen": -290.9193420410156,
621
+ "logps/rejected": -270.82281494140625,
622
+ "loss": 0.6449,
623
+ "positive_losses": 0.07803992927074432,
624
+ "rewards/accuracies": 0.949999988079071,
625
+ "rewards/chosen": 0.1253766566514969,
626
+ "rewards/margins": 0.10763327777385712,
627
+ "rewards/margins_max": 0.20058684051036835,
628
+ "rewards/margins_min": 0.02709307335317135,
629
+ "rewards/margins_std": 0.07942862808704376,
630
+ "rewards/rejected": 0.017743363976478577,
631
+ "step": 280
632
+ },
633
+ {
634
+ "dpo_losses": 0.6451443433761597,
635
+ "epoch": 0.82,
636
+ "grad_norm": 1.741728444197785,
637
+ "learning_rate": 4.563174794266683e-07,
638
+ "logits/chosen": -2.7625350952148438,
639
+ "logits/rejected": -2.7133007049560547,
640
+ "logps/chosen": -275.4216003417969,
641
+ "logps/rejected": -262.64422607421875,
642
+ "loss": 0.655,
643
+ "positive_losses": 0.12835732102394104,
644
+ "rewards/accuracies": 0.8999999761581421,
645
+ "rewards/chosen": 0.11494598537683487,
646
+ "rewards/margins": 0.10120587050914764,
647
+ "rewards/margins_max": 0.21776075661182404,
648
+ "rewards/margins_min": 0.010339265689253807,
649
+ "rewards/margins_std": 0.09577289968729019,
650
+ "rewards/rejected": 0.013740080408751965,
651
+ "step": 290
652
+ },
653
+ {
654
+ "dpo_losses": 0.6270988583564758,
655
+ "epoch": 0.85,
656
+ "grad_norm": 1.8935337862274761,
657
+ "learning_rate": 4.515774809767012e-07,
658
+ "logits/chosen": -2.8255112171173096,
659
+ "logits/rejected": -2.748577833175659,
660
+ "logps/chosen": -331.1516418457031,
661
+ "logps/rejected": -288.11700439453125,
662
+ "loss": 0.6293,
663
+ "positive_losses": 0.02850809134542942,
664
+ "rewards/accuracies": 0.949999988079071,
665
+ "rewards/chosen": 0.1536138951778412,
666
+ "rewards/margins": 0.1404499113559723,
667
+ "rewards/margins_max": 0.27252131700515747,
668
+ "rewards/margins_min": 0.027482062578201294,
669
+ "rewards/margins_std": 0.10937514156103134,
670
+ "rewards/rejected": 0.01316398847848177,
671
+ "step": 300
672
+ },
673
+ {
674
+ "epoch": 0.85,
675
+ "eval_dpo_losses": 0.6802704930305481,
676
+ "eval_logits/chosen": -2.768414258956909,
677
+ "eval_logits/rejected": -2.730605363845825,
678
+ "eval_logps/chosen": -277.8385009765625,
679
+ "eval_logps/rejected": -254.60328674316406,
680
+ "eval_loss": 0.7193401455879211,
681
+ "eval_positive_losses": 0.35523319244384766,
682
+ "eval_rewards/accuracies": 0.5960000157356262,
683
+ "eval_rewards/chosen": 0.06754905730485916,
684
+ "eval_rewards/margins": 0.027792593464255333,
685
+ "eval_rewards/margins_max": 0.16008007526397705,
686
+ "eval_rewards/margins_min": -0.08630798757076263,
687
+ "eval_rewards/margins_std": 0.08258534967899323,
688
+ "eval_rewards/rejected": 0.03975646197795868,
689
+ "eval_runtime": 428.0151,
690
+ "eval_samples_per_second": 4.673,
691
+ "eval_steps_per_second": 0.292,
692
+ "step": 300
693
+ },
694
+ {
695
+ "dpo_losses": 0.6279357671737671,
696
+ "epoch": 0.87,
697
+ "grad_norm": 1.7270459495090738,
698
+ "learning_rate": 4.4662072618099887e-07,
699
+ "logits/chosen": -2.820918560028076,
700
+ "logits/rejected": -2.7266643047332764,
701
+ "logps/chosen": -282.9930725097656,
702
+ "logps/rejected": -213.27236938476562,
703
+ "loss": 0.6267,
704
+ "positive_losses": 0.0,
705
+ "rewards/accuracies": 0.987500011920929,
706
+ "rewards/chosen": 0.14822781085968018,
707
+ "rewards/margins": 0.1387120932340622,
708
+ "rewards/margins_max": 0.27559134364128113,
709
+ "rewards/margins_min": 0.03083965554833412,
710
+ "rewards/margins_std": 0.10963471978902817,
711
+ "rewards/rejected": 0.009515730664134026,
712
+ "step": 310
713
+ },
714
+ {
715
+ "dpo_losses": 0.6319230794906616,
716
+ "epoch": 0.9,
717
+ "grad_norm": 7.736909208406228,
718
+ "learning_rate": 4.414525450399712e-07,
719
+ "logits/chosen": -2.78419828414917,
720
+ "logits/rejected": -2.72727370262146,
721
+ "logps/chosen": -294.29937744140625,
722
+ "logps/rejected": -259.6752014160156,
723
+ "loss": 0.6334,
724
+ "positive_losses": 0.07355575263500214,
725
+ "rewards/accuracies": 0.9375,
726
+ "rewards/chosen": 0.13973799347877502,
727
+ "rewards/margins": 0.12978938221931458,
728
+ "rewards/margins_max": 0.26664164662361145,
729
+ "rewards/margins_min": 0.02372616156935692,
730
+ "rewards/margins_std": 0.10929105430841446,
731
+ "rewards/rejected": 0.009948636405169964,
732
+ "step": 320
733
+ },
734
+ {
735
+ "dpo_losses": 0.6180365085601807,
736
+ "epoch": 0.93,
737
+ "grad_norm": 1.9077047206667683,
738
+ "learning_rate": 4.360784949008615e-07,
739
+ "logits/chosen": -2.8559248447418213,
740
+ "logits/rejected": -2.775313138961792,
741
+ "logps/chosen": -338.38507080078125,
742
+ "logps/rejected": -267.0349426269531,
743
+ "loss": 0.6252,
744
+ "positive_losses": 0.016361523419618607,
745
+ "rewards/accuracies": 0.9624999761581421,
746
+ "rewards/chosen": 0.15934768319129944,
747
+ "rewards/margins": 0.16030099987983704,
748
+ "rewards/margins_max": 0.2862982153892517,
749
+ "rewards/margins_min": 0.039303008466959,
750
+ "rewards/margins_std": 0.11248087882995605,
751
+ "rewards/rejected": -0.0009533234988339245,
752
+ "step": 330
753
+ },
754
+ {
755
+ "dpo_losses": 0.6223454475402832,
756
+ "epoch": 0.96,
757
+ "grad_norm": 1.835838607397251,
758
+ "learning_rate": 4.305043544819289e-07,
759
+ "logits/chosen": -2.741628646850586,
760
+ "logits/rejected": -2.690565586090088,
761
+ "logps/chosen": -283.5008239746094,
762
+ "logps/rejected": -249.989013671875,
763
+ "loss": 0.621,
764
+ "positive_losses": 0.06820764392614365,
765
+ "rewards/accuracies": 0.9375,
766
+ "rewards/chosen": 0.16344675421714783,
767
+ "rewards/margins": 0.15164029598236084,
768
+ "rewards/margins_max": 0.32016128301620483,
769
+ "rewards/margins_min": 0.039167340844869614,
770
+ "rewards/margins_std": 0.12687677145004272,
771
+ "rewards/rejected": 0.011806446127593517,
772
+ "step": 340
773
+ },
774
+ {
775
+ "dpo_losses": 0.6281483173370361,
776
+ "epoch": 0.99,
777
+ "grad_norm": 5.3160905260769695,
778
+ "learning_rate": 4.247361176585903e-07,
779
+ "logits/chosen": -2.7237563133239746,
780
+ "logits/rejected": -2.6812942028045654,
781
+ "logps/chosen": -300.01416015625,
782
+ "logps/rejected": -285.3011779785156,
783
+ "loss": 0.6314,
784
+ "positive_losses": 0.09978675842285156,
785
+ "rewards/accuracies": 0.9375,
786
+ "rewards/chosen": 0.1494494080543518,
787
+ "rewards/margins": 0.1379416286945343,
788
+ "rewards/margins_max": 0.26943984627723694,
789
+ "rewards/margins_min": 0.03328308090567589,
790
+ "rewards/margins_std": 0.10527817159891129,
791
+ "rewards/rejected": 0.011507781222462654,
792
+ "step": 350
793
+ },
794
+ {
795
+ "dpo_losses": 0.6205986142158508,
796
+ "epoch": 1.01,
797
+ "grad_norm": 1.8675515047500797,
798
+ "learning_rate": 4.187799870182038e-07,
799
+ "logits/chosen": -2.76216983795166,
800
+ "logits/rejected": -2.704974412918091,
801
+ "logps/chosen": -284.5892028808594,
802
+ "logps/rejected": -252.9727783203125,
803
+ "loss": 0.6217,
804
+ "positive_losses": 0.02007141150534153,
805
+ "rewards/accuracies": 0.9750000238418579,
806
+ "rewards/chosen": 0.1552152931690216,
807
+ "rewards/margins": 0.15586945414543152,
808
+ "rewards/margins_max": 0.3017035126686096,
809
+ "rewards/margins_min": 0.036038514226675034,
810
+ "rewards/margins_std": 0.11902729421854019,
811
+ "rewards/rejected": -0.0006541303591802716,
812
+ "step": 360
813
+ },
814
+ {
815
+ "dpo_losses": 0.6088902354240417,
816
+ "epoch": 1.04,
817
+ "grad_norm": 1.9531792595002184,
818
+ "learning_rate": 4.126423671904236e-07,
819
+ "logits/chosen": -2.7716732025146484,
820
+ "logits/rejected": -2.7078254222869873,
821
+ "logps/chosen": -336.2352600097656,
822
+ "logps/rejected": -244.32177734375,
823
+ "loss": 0.6125,
824
+ "positive_losses": 0.02638239786028862,
825
+ "rewards/accuracies": 0.949999988079071,
826
+ "rewards/chosen": 0.17914316058158875,
827
+ "rewards/margins": 0.18257828056812286,
828
+ "rewards/margins_max": 0.3468923568725586,
829
+ "rewards/margins_min": 0.04474998638033867,
830
+ "rewards/margins_std": 0.1425740271806717,
831
+ "rewards/rejected": -0.003435105085372925,
832
+ "step": 370
833
+ },
834
+ {
835
+ "dpo_losses": 0.6105533838272095,
836
+ "epoch": 1.07,
837
+ "grad_norm": 1.9479255132083297,
838
+ "learning_rate": 4.0632985796030007e-07,
839
+ "logits/chosen": -2.7388248443603516,
840
+ "logits/rejected": -2.6531803607940674,
841
+ "logps/chosen": -268.4936218261719,
842
+ "logps/rejected": -236.57095336914062,
843
+ "loss": 0.61,
844
+ "positive_losses": 0.008747863583266735,
845
+ "rewards/accuracies": 0.987500011920929,
846
+ "rewards/chosen": 0.1682673990726471,
847
+ "rewards/margins": 0.17755943536758423,
848
+ "rewards/margins_max": 0.3241942226886749,
849
+ "rewards/margins_min": 0.05380522087216377,
850
+ "rewards/margins_std": 0.125107079744339,
851
+ "rewards/rejected": -0.00929203350096941,
852
+ "step": 380
853
+ },
854
+ {
855
+ "dpo_losses": 0.6172813773155212,
856
+ "epoch": 1.1,
857
+ "grad_norm": 8.407134849747322,
858
+ "learning_rate": 3.9984924717152713e-07,
859
+ "logits/chosen": -2.7552475929260254,
860
+ "logits/rejected": -2.7269043922424316,
861
+ "logps/chosen": -285.97760009765625,
862
+ "logps/rejected": -262.98431396484375,
863
+ "loss": 0.6282,
864
+ "positive_losses": 0.06859800964593887,
865
+ "rewards/accuracies": 0.949999988079071,
866
+ "rewards/chosen": 0.15436658263206482,
867
+ "rewards/margins": 0.1631786823272705,
868
+ "rewards/margins_max": 0.3312760889530182,
869
+ "rewards/margins_min": 0.041763290762901306,
870
+ "rewards/margins_std": 0.13355527818202972,
871
+ "rewards/rejected": -0.008812101557850838,
872
+ "step": 390
873
+ },
874
+ {
875
+ "dpo_losses": 0.6174243688583374,
876
+ "epoch": 1.13,
877
+ "grad_norm": 3.651013920669953,
878
+ "learning_rate": 3.932075034274723e-07,
879
+ "logits/chosen": -2.758373975753784,
880
+ "logits/rejected": -2.6981711387634277,
881
+ "logps/chosen": -276.49871826171875,
882
+ "logps/rejected": -245.08224487304688,
883
+ "loss": 0.6236,
884
+ "positive_losses": 0.0625574141740799,
885
+ "rewards/accuracies": 0.925000011920929,
886
+ "rewards/chosen": 0.15731404721736908,
887
+ "rewards/margins": 0.16287846863269806,
888
+ "rewards/margins_max": 0.3262539505958557,
889
+ "rewards/margins_min": 0.043569326400756836,
890
+ "rewards/margins_std": 0.13064977526664734,
891
+ "rewards/rejected": -0.005564402788877487,
892
+ "step": 400
893
+ },
894
+ {
895
+ "epoch": 1.13,
896
+ "eval_dpo_losses": 0.6756274700164795,
897
+ "eval_logits/chosen": -2.749445915222168,
898
+ "eval_logits/rejected": -2.711932420730591,
899
+ "eval_logps/chosen": -276.5967712402344,
900
+ "eval_logps/rejected": -254.45849609375,
901
+ "eval_loss": 0.7519370317459106,
902
+ "eval_positive_losses": 0.6894010305404663,
903
+ "eval_rewards/accuracies": 0.609000027179718,
904
+ "eval_rewards/chosen": 0.07996628433465958,
905
+ "eval_rewards/margins": 0.03876199945807457,
906
+ "eval_rewards/margins_max": 0.21824941039085388,
907
+ "eval_rewards/margins_min": -0.11400224268436432,
908
+ "eval_rewards/margins_std": 0.11133058369159698,
909
+ "eval_rewards/rejected": 0.0412042960524559,
910
+ "eval_runtime": 427.9231,
911
+ "eval_samples_per_second": 4.674,
912
+ "eval_steps_per_second": 0.292,
913
+ "step": 400
914
+ },
915
+ {
916
+ "dpo_losses": 0.6167701482772827,
917
+ "epoch": 1.15,
918
+ "grad_norm": 1.9654856670345158,
919
+ "learning_rate": 3.8641176859783383e-07,
920
+ "logits/chosen": -2.7697155475616455,
921
+ "logits/rejected": -2.707765817642212,
922
+ "logps/chosen": -278.96673583984375,
923
+ "logps/rejected": -248.4454345703125,
924
+ "loss": 0.6213,
925
+ "positive_losses": 0.21929892897605896,
926
+ "rewards/accuracies": 0.9125000238418579,
927
+ "rewards/chosen": 0.17709851264953613,
928
+ "rewards/margins": 0.16643425822257996,
929
+ "rewards/margins_max": 0.3442317843437195,
930
+ "rewards/margins_min": 0.015078430995345116,
931
+ "rewards/margins_std": 0.15533670783042908,
932
+ "rewards/rejected": 0.010664242319762707,
933
+ "step": 410
934
+ },
935
+ {
936
+ "dpo_losses": 0.6068592071533203,
937
+ "epoch": 1.18,
938
+ "grad_norm": 2.3100164325344315,
939
+ "learning_rate": 3.7946935013898606e-07,
940
+ "logits/chosen": -2.7368392944335938,
941
+ "logits/rejected": -2.677018642425537,
942
+ "logps/chosen": -298.12542724609375,
943
+ "logps/rejected": -266.35406494140625,
944
+ "loss": 0.6085,
945
+ "positive_losses": 0.038385771214962006,
946
+ "rewards/accuracies": 0.9375,
947
+ "rewards/chosen": 0.19851230084896088,
948
+ "rewards/margins": 0.18865473568439484,
949
+ "rewards/margins_max": 0.4086156487464905,
950
+ "rewards/margins_min": 0.0322859063744545,
951
+ "rewards/margins_std": 0.16774147748947144,
952
+ "rewards/rejected": 0.00985757913440466,
953
+ "step": 420
954
+ },
955
+ {
956
+ "dpo_losses": 0.6006068587303162,
957
+ "epoch": 1.21,
958
+ "grad_norm": 6.525192737290529,
959
+ "learning_rate": 3.7238771323626817e-07,
960
+ "logits/chosen": -2.768768787384033,
961
+ "logits/rejected": -2.681751012802124,
962
+ "logps/chosen": -323.71514892578125,
963
+ "logps/rejected": -260.93621826171875,
964
+ "loss": 0.6082,
965
+ "positive_losses": 0.09947166591882706,
966
+ "rewards/accuracies": 0.9375,
967
+ "rewards/chosen": 0.1950441300868988,
968
+ "rewards/margins": 0.2026427984237671,
969
+ "rewards/margins_max": 0.3679996430873871,
970
+ "rewards/margins_min": 0.06226770952343941,
971
+ "rewards/margins_std": 0.1420767605304718,
972
+ "rewards/rejected": -0.007598669268190861,
973
+ "step": 430
974
+ },
975
+ {
976
+ "dpo_losses": 0.6032605171203613,
977
+ "epoch": 1.24,
978
+ "grad_norm": 1.6385345947087144,
979
+ "learning_rate": 3.651744727766676e-07,
980
+ "logits/chosen": -2.768833637237549,
981
+ "logits/rejected": -2.702613353729248,
982
+ "logps/chosen": -273.583740234375,
983
+ "logps/rejected": -246.775390625,
984
+ "loss": 0.6049,
985
+ "positive_losses": 0.0,
986
+ "rewards/accuracies": 0.9624999761581421,
987
+ "rewards/chosen": 0.17607031762599945,
988
+ "rewards/margins": 0.1937752068042755,
989
+ "rewards/margins_max": 0.34389567375183105,
990
+ "rewards/margins_min": 0.06312290579080582,
991
+ "rewards/margins_std": 0.1278633177280426,
992
+ "rewards/rejected": -0.017704878002405167,
993
+ "step": 440
994
+ },
995
+ {
996
+ "dpo_losses": 0.6007322072982788,
997
+ "epoch": 1.27,
998
+ "grad_norm": 1.9767106796026328,
999
+ "learning_rate": 3.5783738516052897e-07,
1000
+ "logits/chosen": -2.7184462547302246,
1001
+ "logits/rejected": -2.658703327178955,
1002
+ "logps/chosen": -294.31878662109375,
1003
+ "logps/rejected": -261.46221923828125,
1004
+ "loss": 0.6054,
1005
+ "positive_losses": 0.05530524253845215,
1006
+ "rewards/accuracies": 1.0,
1007
+ "rewards/chosen": 0.17491844296455383,
1008
+ "rewards/margins": 0.20017877221107483,
1009
+ "rewards/margins_max": 0.37646952271461487,
1010
+ "rewards/margins_min": 0.05014311149716377,
1011
+ "rewards/margins_std": 0.14955812692642212,
1012
+ "rewards/rejected": -0.02526034787297249,
1013
+ "step": 450
1014
+ },
1015
+ {
1016
+ "dpo_losses": 0.5913182497024536,
1017
+ "epoch": 1.3,
1018
+ "grad_norm": 1.9303555587009904,
1019
+ "learning_rate": 3.5038433996109404e-07,
1020
+ "logits/chosen": -2.786879777908325,
1021
+ "logits/rejected": -2.722357749938965,
1022
+ "logps/chosen": -335.79766845703125,
1023
+ "logps/rejected": -299.84027099609375,
1024
+ "loss": 0.5936,
1025
+ "positive_losses": 0.03368115425109863,
1026
+ "rewards/accuracies": 0.987500011920929,
1027
+ "rewards/chosen": 0.19536344707012177,
1028
+ "rewards/margins": 0.22316019237041473,
1029
+ "rewards/margins_max": 0.4146662652492523,
1030
+ "rewards/margins_min": 0.05830659344792366,
1031
+ "rewards/margins_std": 0.1629330813884735,
1032
+ "rewards/rejected": -0.027796756476163864,
1033
+ "step": 460
1034
+ },
1035
+ {
1036
+ "dpo_losses": 0.5880565643310547,
1037
+ "epoch": 1.32,
1038
+ "grad_norm": 5.289024274003296,
1039
+ "learning_rate": 3.428233514408398e-07,
1040
+ "logits/chosen": -2.7258143424987793,
1041
+ "logits/rejected": -2.703859806060791,
1042
+ "logps/chosen": -296.7102355957031,
1043
+ "logps/rejected": -250.7404022216797,
1044
+ "loss": 0.6047,
1045
+ "positive_losses": 0.15512129664421082,
1046
+ "rewards/accuracies": 0.9750000238418579,
1047
+ "rewards/chosen": 0.19506558775901794,
1048
+ "rewards/margins": 0.23098190128803253,
1049
+ "rewards/margins_max": 0.4174633026123047,
1050
+ "rewards/margins_min": 0.0681539997458458,
1051
+ "rewards/margins_std": 0.15801438689231873,
1052
+ "rewards/rejected": -0.0359162911772728,
1053
+ "step": 470
1054
+ },
1055
+ {
1056
+ "dpo_losses": 0.5843189358711243,
1057
+ "epoch": 1.35,
1058
+ "grad_norm": 1.5003059393102265,
1059
+ "learning_rate": 3.3516254993373945e-07,
1060
+ "logits/chosen": -2.755491018295288,
1061
+ "logits/rejected": -2.681509256362915,
1062
+ "logps/chosen": -311.5091247558594,
1063
+ "logps/rejected": -245.88937377929688,
1064
+ "loss": 0.5939,
1065
+ "positive_losses": 0.055457282811403275,
1066
+ "rewards/accuracies": 0.987500011920929,
1067
+ "rewards/chosen": 0.19728389382362366,
1068
+ "rewards/margins": 0.23806998133659363,
1069
+ "rewards/margins_max": 0.4085385203361511,
1070
+ "rewards/margins_min": 0.06948045641183853,
1071
+ "rewards/margins_std": 0.15313410758972168,
1072
+ "rewards/rejected": -0.04078609496355057,
1073
+ "step": 480
1074
+ },
1075
+ {
1076
+ "dpo_losses": 0.588597297668457,
1077
+ "epoch": 1.38,
1078
+ "grad_norm": 2.2336443793585277,
1079
+ "learning_rate": 3.274101731027105e-07,
1080
+ "logits/chosen": -2.709615707397461,
1081
+ "logits/rejected": -2.663390636444092,
1082
+ "logps/chosen": -318.6739807128906,
1083
+ "logps/rejected": -286.79541015625,
1084
+ "loss": 0.593,
1085
+ "positive_losses": 0.0,
1086
+ "rewards/accuracies": 0.9375,
1087
+ "rewards/chosen": 0.20303764939308167,
1088
+ "rewards/margins": 0.23158419132232666,
1089
+ "rewards/margins_max": 0.457578182220459,
1090
+ "rewards/margins_min": 0.04808598756790161,
1091
+ "rewards/margins_std": 0.18310314416885376,
1092
+ "rewards/rejected": -0.028546560555696487,
1093
+ "step": 490
1094
+ },
1095
+ {
1096
+ "dpo_losses": 0.598129153251648,
1097
+ "epoch": 1.41,
1098
+ "grad_norm": 1.719544868467305,
1099
+ "learning_rate": 3.1957455708165314e-07,
1100
+ "logits/chosen": -2.6921727657318115,
1101
+ "logits/rejected": -2.678170680999756,
1102
+ "logps/chosen": -280.46063232421875,
1103
+ "logps/rejected": -249.769287109375,
1104
+ "loss": 0.6009,
1105
+ "positive_losses": 0.036452483385801315,
1106
+ "rewards/accuracies": 0.949999988079071,
1107
+ "rewards/chosen": 0.17304545640945435,
1108
+ "rewards/margins": 0.2069230079650879,
1109
+ "rewards/margins_max": 0.40413007140159607,
1110
+ "rewards/margins_min": 0.054103028029203415,
1111
+ "rewards/margins_std": 0.15941500663757324,
1112
+ "rewards/rejected": -0.03387756645679474,
1113
+ "step": 500
1114
+ },
1115
+ {
1116
+ "epoch": 1.41,
1117
+ "eval_dpo_losses": 0.6718186140060425,
1118
+ "eval_logits/chosen": -2.7294745445251465,
1119
+ "eval_logits/rejected": -2.692014694213867,
1120
+ "eval_logps/chosen": -278.2060852050781,
1121
+ "eval_logps/rejected": -257.0343322753906,
1122
+ "eval_loss": 0.8434445858001709,
1123
+ "eval_positive_losses": 1.5495002269744873,
1124
+ "eval_rewards/accuracies": 0.609000027179718,
1125
+ "eval_rewards/chosen": 0.06387308984994888,
1126
+ "eval_rewards/margins": 0.04842698201537132,
1127
+ "eval_rewards/margins_max": 0.2708706855773926,
1128
+ "eval_rewards/margins_min": -0.1440476030111313,
1129
+ "eval_rewards/margins_std": 0.1388508826494217,
1130
+ "eval_rewards/rejected": 0.015446108765900135,
1131
+ "eval_runtime": 427.8728,
1132
+ "eval_samples_per_second": 4.674,
1133
+ "eval_steps_per_second": 0.292,
1134
+ "step": 500
1135
+ },
1136
+ {
1137
+ "dpo_losses": 0.5843192338943481,
1138
+ "epoch": 1.44,
1139
+ "grad_norm": 3.509981635934115,
1140
+ "learning_rate": 3.116641275116018e-07,
1141
+ "logits/chosen": -2.7144105434417725,
1142
+ "logits/rejected": -2.628718614578247,
1143
+ "logps/chosen": -321.0252380371094,
1144
+ "logps/rejected": -263.8638000488281,
1145
+ "loss": 0.5918,
1146
+ "positive_losses": 0.06754092872142792,
1147
+ "rewards/accuracies": 0.9750000238418579,
1148
+ "rewards/chosen": 0.1961548924446106,
1149
+ "rewards/margins": 0.24090898036956787,
1150
+ "rewards/margins_max": 0.46518999338150024,
1151
+ "rewards/margins_min": 0.07308268547058105,
1152
+ "rewards/margins_std": 0.1766786277294159,
1153
+ "rewards/rejected": -0.04475412517786026,
1154
+ "step": 510
1155
+ },
1156
+ {
1157
+ "dpo_losses": 0.5933550596237183,
1158
+ "epoch": 1.46,
1159
+ "grad_norm": 1.7887167117515541,
1160
+ "learning_rate": 3.036873904806295e-07,
1161
+ "logits/chosen": -2.7657430171966553,
1162
+ "logits/rejected": -2.6845974922180176,
1163
+ "logps/chosen": -281.34649658203125,
1164
+ "logps/rejected": -241.7858123779297,
1165
+ "loss": 0.6137,
1166
+ "positive_losses": 0.2598763406276703,
1167
+ "rewards/accuracies": 0.9624999761581421,
1168
+ "rewards/chosen": 0.19071096181869507,
1169
+ "rewards/margins": 0.22014352679252625,
1170
+ "rewards/margins_max": 0.44077128171920776,
1171
+ "rewards/margins_min": 0.0637890174984932,
1172
+ "rewards/margins_std": 0.17364180088043213,
1173
+ "rewards/rejected": -0.02943255938589573,
1174
+ "step": 520
1175
+ },
1176
+ {
1177
+ "dpo_losses": 0.5821532011032104,
1178
+ "epoch": 1.49,
1179
+ "grad_norm": 10.11338724174373,
1180
+ "learning_rate": 2.956529233772492e-07,
1181
+ "logits/chosen": -2.769479513168335,
1182
+ "logits/rejected": -2.7147421836853027,
1183
+ "logps/chosen": -314.7546081542969,
1184
+ "logps/rejected": -300.4794921875,
1185
+ "loss": 0.618,
1186
+ "positive_losses": 0.35480666160583496,
1187
+ "rewards/accuracies": 0.9624999761581421,
1188
+ "rewards/chosen": 0.19147683680057526,
1189
+ "rewards/margins": 0.2443142831325531,
1190
+ "rewards/margins_max": 0.41471806168556213,
1191
+ "rewards/margins_min": 0.07710012048482895,
1192
+ "rewards/margins_std": 0.14999482035636902,
1193
+ "rewards/rejected": -0.05283746123313904,
1194
+ "step": 530
1195
+ },
1196
+ {
1197
+ "dpo_losses": 0.5760180950164795,
1198
+ "epoch": 1.52,
1199
+ "grad_norm": 1.9182115700216993,
1200
+ "learning_rate": 2.875693656671431e-07,
1201
+ "logits/chosen": -2.761521577835083,
1202
+ "logits/rejected": -2.678863286972046,
1203
+ "logps/chosen": -320.120849609375,
1204
+ "logps/rejected": -249.3918914794922,
1205
+ "loss": 0.5887,
1206
+ "positive_losses": 0.12351331859827042,
1207
+ "rewards/accuracies": 0.949999988079071,
1208
+ "rewards/chosen": 0.2093885838985443,
1209
+ "rewards/margins": 0.26241612434387207,
1210
+ "rewards/margins_max": 0.5384218692779541,
1211
+ "rewards/margins_min": 0.05588661879301071,
1212
+ "rewards/margins_std": 0.2135990560054779,
1213
+ "rewards/rejected": -0.05302751809358597,
1214
+ "step": 540
1215
+ },
1216
+ {
1217
+ "dpo_losses": 0.5680991411209106,
1218
+ "epoch": 1.55,
1219
+ "grad_norm": 1.947863946973032,
1220
+ "learning_rate": 2.794454096031429e-07,
1221
+ "logits/chosen": -2.7631947994232178,
1222
+ "logits/rejected": -2.695374011993408,
1223
+ "logps/chosen": -349.5855407714844,
1224
+ "logps/rejected": -319.5202941894531,
1225
+ "loss": 0.5909,
1226
+ "positive_losses": 0.10299448668956757,
1227
+ "rewards/accuracies": 0.9624999761581421,
1228
+ "rewards/chosen": 0.23324398696422577,
1229
+ "rewards/margins": 0.2784574627876282,
1230
+ "rewards/margins_max": 0.5278440713882446,
1231
+ "rewards/margins_min": 0.08051635324954987,
1232
+ "rewards/margins_std": 0.20143046975135803,
1233
+ "rewards/rejected": -0.04521343857049942,
1234
+ "step": 550
1235
+ },
1236
+ {
1237
+ "dpo_losses": 0.5882316827774048,
1238
+ "epoch": 1.58,
1239
+ "grad_norm": 6.883043057999946,
1240
+ "learning_rate": 2.7128979087844593e-07,
1241
+ "logits/chosen": -2.733025312423706,
1242
+ "logits/rejected": -2.6712567806243896,
1243
+ "logps/chosen": -285.25115966796875,
1244
+ "logps/rejected": -269.1498107910156,
1245
+ "loss": 0.5847,
1246
+ "positive_losses": 0.1969757080078125,
1247
+ "rewards/accuracies": 0.949999988079071,
1248
+ "rewards/chosen": 0.1997162401676178,
1249
+ "rewards/margins": 0.23153042793273926,
1250
+ "rewards/margins_max": 0.4535094201564789,
1251
+ "rewards/margins_min": 0.05927438288927078,
1252
+ "rewards/margins_std": 0.17876756191253662,
1253
+ "rewards/rejected": -0.03181419149041176,
1254
+ "step": 560
1255
+ },
1256
+ {
1257
+ "dpo_losses": 0.5802090764045715,
1258
+ "epoch": 1.61,
1259
+ "grad_norm": 1.7758243312216686,
1260
+ "learning_rate": 2.6311127923312153e-07,
1261
+ "logits/chosen": -2.7622435092926025,
1262
+ "logits/rejected": -2.6936147212982178,
1263
+ "logps/chosen": -315.955322265625,
1264
+ "logps/rejected": -297.02313232421875,
1265
+ "loss": 0.5892,
1266
+ "positive_losses": 0.21824970841407776,
1267
+ "rewards/accuracies": 0.9750000238418579,
1268
+ "rewards/chosen": 0.1967122107744217,
1269
+ "rewards/margins": 0.2509615123271942,
1270
+ "rewards/margins_max": 0.4743427336215973,
1271
+ "rewards/margins_min": 0.049542300403118134,
1272
+ "rewards/margins_std": 0.19265086948871613,
1273
+ "rewards/rejected": -0.05424932390451431,
1274
+ "step": 570
1275
+ },
1276
+ {
1277
+ "dpo_losses": 0.5920171737670898,
1278
+ "epoch": 1.63,
1279
+ "grad_norm": 1.860643226808477,
1280
+ "learning_rate": 2.5491866902400565e-07,
1281
+ "logits/chosen": -2.7078280448913574,
1282
+ "logits/rejected": -2.6779098510742188,
1283
+ "logps/chosen": -269.2068786621094,
1284
+ "logps/rejected": -271.0436096191406,
1285
+ "loss": 0.5991,
1286
+ "positive_losses": 0.12158451229333878,
1287
+ "rewards/accuracies": 0.9750000238418579,
1288
+ "rewards/chosen": 0.17235751450061798,
1289
+ "rewards/margins": 0.22299370169639587,
1290
+ "rewards/margins_max": 0.43409761786460876,
1291
+ "rewards/margins_min": 0.05049045756459236,
1292
+ "rewards/margins_std": 0.1718008816242218,
1293
+ "rewards/rejected": -0.05063622072339058,
1294
+ "step": 580
1295
+ },
1296
+ {
1297
+ "dpo_losses": 0.5816795229911804,
1298
+ "epoch": 1.66,
1299
+ "grad_norm": 24.09510524317517,
1300
+ "learning_rate": 2.4672076976812543e-07,
1301
+ "logits/chosen": -2.720360279083252,
1302
+ "logits/rejected": -2.6442887783050537,
1303
+ "logps/chosen": -288.3584289550781,
1304
+ "logps/rejected": -267.7664794921875,
1305
+ "loss": 0.5961,
1306
+ "positive_losses": 0.3148724436759949,
1307
+ "rewards/accuracies": 0.9375,
1308
+ "rewards/chosen": 0.183799147605896,
1309
+ "rewards/margins": 0.24748222529888153,
1310
+ "rewards/margins_max": 0.4496613144874573,
1311
+ "rewards/margins_min": 0.06704072654247284,
1312
+ "rewards/margins_std": 0.17722192406654358,
1313
+ "rewards/rejected": -0.06368308514356613,
1314
+ "step": 590
1315
+ },
1316
+ {
1317
+ "dpo_losses": 0.576720118522644,
1318
+ "epoch": 1.69,
1319
+ "grad_norm": 2.1497917207479507,
1320
+ "learning_rate": 2.385263966698222e-07,
1321
+ "logits/chosen": -2.728609085083008,
1322
+ "logits/rejected": -2.6675384044647217,
1323
+ "logps/chosen": -273.47369384765625,
1324
+ "logps/rejected": -296.99310302734375,
1325
+ "loss": 0.6136,
1326
+ "positive_losses": 0.32597580552101135,
1327
+ "rewards/accuracies": 0.949999988079071,
1328
+ "rewards/chosen": 0.19901221990585327,
1329
+ "rewards/margins": 0.26053741574287415,
1330
+ "rewards/margins_max": 0.5135586261749268,
1331
+ "rewards/margins_min": 0.06456376612186432,
1332
+ "rewards/margins_std": 0.20331530272960663,
1333
+ "rewards/rejected": -0.06152517348527908,
1334
+ "step": 600
1335
+ },
1336
+ {
1337
+ "epoch": 1.69,
1338
+ "eval_dpo_losses": 0.6691358685493469,
1339
+ "eval_logits/chosen": -2.7203145027160645,
1340
+ "eval_logits/rejected": -2.682708263397217,
1341
+ "eval_logps/chosen": -277.7243957519531,
1342
+ "eval_logps/rejected": -257.23602294921875,
1343
+ "eval_loss": 0.8727347254753113,
1344
+ "eval_positive_losses": 1.8302409648895264,
1345
+ "eval_rewards/accuracies": 0.6129999756813049,
1346
+ "eval_rewards/chosen": 0.06869003176689148,
1347
+ "eval_rewards/margins": 0.055261291563510895,
1348
+ "eval_rewards/margins_max": 0.30490291118621826,
1349
+ "eval_rewards/margins_min": -0.15945661067962646,
1350
+ "eval_rewards/margins_std": 0.1552800089120865,
1351
+ "eval_rewards/rejected": 0.013428742997348309,
1352
+ "eval_runtime": 427.6004,
1353
+ "eval_samples_per_second": 4.677,
1354
+ "eval_steps_per_second": 0.292,
1355
+ "step": 600
1356
+ },
1357
+ {
1358
+ "dpo_losses": 0.5747733116149902,
1359
+ "epoch": 1.72,
1360
+ "grad_norm": 6.65096965403295,
1361
+ "learning_rate": 2.3034436114175838e-07,
1362
+ "logits/chosen": -2.7278988361358643,
1363
+ "logits/rejected": -2.6467108726501465,
1364
+ "logps/chosen": -305.2555847167969,
1365
+ "logps/rejected": -246.3316192626953,
1366
+ "loss": 0.5995,
1367
+ "positive_losses": 0.20225219428539276,
1368
+ "rewards/accuracies": 0.925000011920929,
1369
+ "rewards/chosen": 0.22444753348827362,
1370
+ "rewards/margins": 0.2688870429992676,
1371
+ "rewards/margins_max": 0.5200773477554321,
1372
+ "rewards/margins_min": 0.04230537265539169,
1373
+ "rewards/margins_std": 0.21579697728157043,
1374
+ "rewards/rejected": -0.044439516961574554,
1375
+ "step": 610
1376
+ },
1377
+ {
1378
+ "dpo_losses": 0.5641311407089233,
1379
+ "epoch": 1.75,
1380
+ "grad_norm": 11.81079993553305,
1381
+ "learning_rate": 2.2218346133000264e-07,
1382
+ "logits/chosen": -2.7867112159729004,
1383
+ "logits/rejected": -2.71598744392395,
1384
+ "logps/chosen": -318.68121337890625,
1385
+ "logps/rejected": -289.48309326171875,
1386
+ "loss": 0.6115,
1387
+ "positive_losses": 0.15610671043395996,
1388
+ "rewards/accuracies": 0.9624999761581421,
1389
+ "rewards/chosen": 0.22943755984306335,
1390
+ "rewards/margins": 0.29184073209762573,
1391
+ "rewards/margins_max": 0.5524952411651611,
1392
+ "rewards/margins_min": 0.07403835654258728,
1393
+ "rewards/margins_std": 0.2165592461824417,
1394
+ "rewards/rejected": -0.06240314990282059,
1395
+ "step": 620
1396
+ },
1397
+ {
1398
+ "dpo_losses": 0.5642560720443726,
1399
+ "epoch": 1.77,
1400
+ "grad_norm": 7.357647930061899,
1401
+ "learning_rate": 2.1405247265337917e-07,
1402
+ "logits/chosen": -2.693892002105713,
1403
+ "logits/rejected": -2.6591708660125732,
1404
+ "logps/chosen": -349.5628356933594,
1405
+ "logps/rejected": -286.9067687988281,
1406
+ "loss": 0.5786,
1407
+ "positive_losses": 0.0,
1408
+ "rewards/accuracies": 0.9750000238418579,
1409
+ "rewards/chosen": 0.2449304610490799,
1410
+ "rewards/margins": 0.28937476873397827,
1411
+ "rewards/margins_max": 0.5107169151306152,
1412
+ "rewards/margins_min": 0.09730032831430435,
1413
+ "rewards/margins_std": 0.18998651206493378,
1414
+ "rewards/rejected": -0.04444430395960808,
1415
+ "step": 630
1416
+ },
1417
+ {
1418
+ "dpo_losses": 0.5737847089767456,
1419
+ "epoch": 1.8,
1420
+ "grad_norm": 1.3349176272864234,
1421
+ "learning_rate": 2.0596013836725657e-07,
1422
+ "logits/chosen": -2.779600143432617,
1423
+ "logits/rejected": -2.700307846069336,
1424
+ "logps/chosen": -290.4892578125,
1425
+ "logps/rejected": -238.4832305908203,
1426
+ "loss": 0.5953,
1427
+ "positive_losses": 0.20754241943359375,
1428
+ "rewards/accuracies": 0.925000011920929,
1429
+ "rewards/chosen": 0.2137231081724167,
1430
+ "rewards/margins": 0.2684459686279297,
1431
+ "rewards/margins_max": 0.4815581440925598,
1432
+ "rewards/margins_min": 0.08431941270828247,
1433
+ "rewards/margins_std": 0.1738905906677246,
1434
+ "rewards/rejected": -0.0547228567302227,
1435
+ "step": 640
1436
+ },
1437
+ {
1438
+ "dpo_losses": 0.5745794773101807,
1439
+ "epoch": 1.83,
1440
+ "grad_norm": 2.223670875381027,
1441
+ "learning_rate": 1.9791516016192213e-07,
1442
+ "logits/chosen": -2.7533204555511475,
1443
+ "logits/rejected": -2.6931979656219482,
1444
+ "logps/chosen": -308.6822814941406,
1445
+ "logps/rejected": -296.13787841796875,
1446
+ "loss": 0.6091,
1447
+ "positive_losses": 0.41213884949684143,
1448
+ "rewards/accuracies": 0.925000011920929,
1449
+ "rewards/chosen": 0.21507827937602997,
1450
+ "rewards/margins": 0.26380831003189087,
1451
+ "rewards/margins_max": 0.4691304564476013,
1452
+ "rewards/margins_min": 0.051321595907211304,
1453
+ "rewards/margins_std": 0.18508048355579376,
1454
+ "rewards/rejected": -0.04873000457882881,
1455
+ "step": 650
1456
+ },
1457
+ {
1458
+ "dpo_losses": 0.5978478193283081,
1459
+ "epoch": 1.86,
1460
+ "grad_norm": 1.9229193850326198,
1461
+ "learning_rate": 1.8992618880565036e-07,
1462
+ "logits/chosen": -2.676647663116455,
1463
+ "logits/rejected": -2.6538872718811035,
1464
+ "logps/chosen": -228.5106658935547,
1465
+ "logps/rejected": -218.3628387451172,
1466
+ "loss": 0.6228,
1467
+ "positive_losses": 0.36836346983909607,
1468
+ "rewards/accuracies": 0.925000011920929,
1469
+ "rewards/chosen": 0.17657354474067688,
1470
+ "rewards/margins": 0.20908644795417786,
1471
+ "rewards/margins_max": 0.38408344984054565,
1472
+ "rewards/margins_min": 0.0495840385556221,
1473
+ "rewards/margins_std": 0.1525198072195053,
1474
+ "rewards/rejected": -0.03251289203763008,
1475
+ "step": 660
1476
+ },
1477
+ {
1478
+ "dpo_losses": 0.5676581859588623,
1479
+ "epoch": 1.89,
1480
+ "grad_norm": 10.31975804008308,
1481
+ "learning_rate": 1.8200181484252885e-07,
1482
+ "logits/chosen": -2.772998809814453,
1483
+ "logits/rejected": -2.742371082305908,
1484
+ "logps/chosen": -282.56695556640625,
1485
+ "logps/rejected": -246.2154083251953,
1486
+ "loss": 0.5836,
1487
+ "positive_losses": 0.19931070506572723,
1488
+ "rewards/accuracies": 1.0,
1489
+ "rewards/chosen": 0.23336899280548096,
1490
+ "rewards/margins": 0.28042665123939514,
1491
+ "rewards/margins_max": 0.5176241397857666,
1492
+ "rewards/margins_min": 0.08305230736732483,
1493
+ "rewards/margins_std": 0.19908221065998077,
1494
+ "rewards/rejected": -0.047057636082172394,
1495
+ "step": 670
1496
+ },
1497
+ {
1498
+ "dpo_losses": 0.570050835609436,
1499
+ "epoch": 1.92,
1500
+ "grad_norm": 6.2395285814671375,
1501
+ "learning_rate": 1.7415055935504233e-07,
1502
+ "logits/chosen": -2.791836738586426,
1503
+ "logits/rejected": -2.7229678630828857,
1504
+ "logps/chosen": -289.1046142578125,
1505
+ "logps/rejected": -257.9730224609375,
1506
+ "loss": 0.5814,
1507
+ "positive_losses": 0.15246662497520447,
1508
+ "rewards/accuracies": 0.9375,
1509
+ "rewards/chosen": 0.21605853736400604,
1510
+ "rewards/margins": 0.274807333946228,
1511
+ "rewards/margins_max": 0.46929216384887695,
1512
+ "rewards/margins_min": 0.07633324712514877,
1513
+ "rewards/margins_std": 0.17314451932907104,
1514
+ "rewards/rejected": -0.05874878168106079,
1515
+ "step": 680
1516
+ },
1517
+ {
1518
+ "dpo_losses": 0.5896421074867249,
1519
+ "epoch": 1.94,
1520
+ "grad_norm": 9.413363887459912,
1521
+ "learning_rate": 1.6638086480134952e-07,
1522
+ "logits/chosen": -2.7295353412628174,
1523
+ "logits/rejected": -2.695042133331299,
1524
+ "logps/chosen": -219.7963409423828,
1525
+ "logps/rejected": -206.1276397705078,
1526
+ "loss": 0.5879,
1527
+ "positive_losses": 0.11815419048070908,
1528
+ "rewards/accuracies": 0.9375,
1529
+ "rewards/chosen": 0.18556587398052216,
1530
+ "rewards/margins": 0.22936753928661346,
1531
+ "rewards/margins_max": 0.44191575050354004,
1532
+ "rewards/margins_min": 0.05077395588159561,
1533
+ "rewards/margins_std": 0.17514900863170624,
1534
+ "rewards/rejected": -0.0438016876578331,
1535
+ "step": 690
1536
+ },
1537
+ {
1538
+ "dpo_losses": 0.5733457207679749,
1539
+ "epoch": 1.97,
1540
+ "grad_norm": 5.98740527370756,
1541
+ "learning_rate": 1.5870108593710471e-07,
1542
+ "logits/chosen": -2.6704626083374023,
1543
+ "logits/rejected": -2.5912749767303467,
1544
+ "logps/chosen": -277.55157470703125,
1545
+ "logps/rejected": -214.6815948486328,
1546
+ "loss": 0.5918,
1547
+ "positive_losses": 0.17365257441997528,
1548
+ "rewards/accuracies": 0.9624999761581421,
1549
+ "rewards/chosen": 0.21979399025440216,
1550
+ "rewards/margins": 0.26772385835647583,
1551
+ "rewards/margins_max": 0.4699093699455261,
1552
+ "rewards/margins_min": 0.09799458831548691,
1553
+ "rewards/margins_std": 0.17043128609657288,
1554
+ "rewards/rejected": -0.047929856926202774,
1555
+ "step": 700
1556
+ },
1557
+ {
1558
+ "epoch": 1.97,
1559
+ "eval_dpo_losses": 0.6676550507545471,
1560
+ "eval_logits/chosen": -2.7172420024871826,
1561
+ "eval_logits/rejected": -2.679697036743164,
1562
+ "eval_logps/chosen": -277.88079833984375,
1563
+ "eval_logps/rejected": -257.7734069824219,
1564
+ "eval_loss": 0.8997932076454163,
1565
+ "eval_positive_losses": 2.0811269283294678,
1566
+ "eval_rewards/accuracies": 0.621999979019165,
1567
+ "eval_rewards/chosen": 0.06712605804204941,
1568
+ "eval_rewards/margins": 0.0590706393122673,
1569
+ "eval_rewards/margins_max": 0.32306957244873047,
1570
+ "eval_rewards/margins_min": -0.16852563619613647,
1571
+ "eval_rewards/margins_std": 0.1640576869249344,
1572
+ "eval_rewards/rejected": 0.00805541779845953,
1573
+ "eval_runtime": 427.6887,
1574
+ "eval_samples_per_second": 4.676,
1575
+ "eval_steps_per_second": 0.292,
1576
+ "step": 700
1577
+ },
1578
+ {
1579
+ "dpo_losses": 0.5846664309501648,
1580
+ "epoch": 2.0,
1581
+ "grad_norm": 2.012900588641833,
1582
+ "learning_rate": 1.5111948083158528e-07,
1583
+ "logits/chosen": -2.665996551513672,
1584
+ "logits/rejected": -2.6332955360412598,
1585
+ "logps/chosen": -233.13113403320312,
1586
+ "logps/rejected": -231.0898895263672,
1587
+ "loss": 0.5777,
1588
+ "positive_losses": 0.08777942508459091,
1589
+ "rewards/accuracies": 0.9624999761581421,
1590
+ "rewards/chosen": 0.17990438640117645,
1591
+ "rewards/margins": 0.23856917023658752,
1592
+ "rewards/margins_max": 0.4145428240299225,
1593
+ "rewards/margins_min": 0.06277941167354584,
1594
+ "rewards/margins_std": 0.15543320775032043,
1595
+ "rewards/rejected": -0.058664750307798386,
1596
+ "step": 710
1597
+ },
1598
+ {
1599
+ "dpo_losses": 0.5619566440582275,
1600
+ "epoch": 2.03,
1601
+ "grad_norm": 8.927999929393918,
1602
+ "learning_rate": 1.4364420198778658e-07,
1603
+ "logits/chosen": -2.803697109222412,
1604
+ "logits/rejected": -2.725480556488037,
1605
+ "logps/chosen": -341.5400085449219,
1606
+ "logps/rejected": -310.74261474609375,
1607
+ "loss": 0.5941,
1608
+ "positive_losses": 0.07888917624950409,
1609
+ "rewards/accuracies": 0.949999988079071,
1610
+ "rewards/chosen": 0.2251276969909668,
1611
+ "rewards/margins": 0.2960202395915985,
1612
+ "rewards/margins_max": 0.5476168394088745,
1613
+ "rewards/margins_min": 0.08313913643360138,
1614
+ "rewards/margins_std": 0.21061666309833527,
1615
+ "rewards/rejected": -0.0708925798535347,
1616
+ "step": 720
1617
+ },
1618
+ {
1619
+ "dpo_losses": 0.5583115816116333,
1620
+ "epoch": 2.06,
1621
+ "grad_norm": 1.8517093519732466,
1622
+ "learning_rate": 1.3628328757603242e-07,
1623
+ "logits/chosen": -2.7375550270080566,
1624
+ "logits/rejected": -2.673107862472534,
1625
+ "logps/chosen": -314.8241271972656,
1626
+ "logps/rejected": -281.47967529296875,
1627
+ "loss": 0.5671,
1628
+ "positive_losses": 0.10358180850744247,
1629
+ "rewards/accuracies": 0.949999988079071,
1630
+ "rewards/chosen": 0.23430731892585754,
1631
+ "rewards/margins": 0.3044988512992859,
1632
+ "rewards/margins_max": 0.5319117307662964,
1633
+ "rewards/margins_min": 0.0928182452917099,
1634
+ "rewards/margins_std": 0.199130579829216,
1635
+ "rewards/rejected": -0.07019157707691193,
1636
+ "step": 730
1637
+ },
1638
+ {
1639
+ "dpo_losses": 0.5659884810447693,
1640
+ "epoch": 2.08,
1641
+ "grad_norm": 9.541706290572293,
1642
+ "learning_rate": 1.2904465279052723e-07,
1643
+ "logits/chosen": -2.7077741622924805,
1644
+ "logits/rejected": -2.6461472511291504,
1645
+ "logps/chosen": -299.73486328125,
1646
+ "logps/rejected": -277.62725830078125,
1647
+ "loss": 0.6025,
1648
+ "positive_losses": 0.5914154052734375,
1649
+ "rewards/accuracies": 0.9750000238418579,
1650
+ "rewards/chosen": 0.20683905482292175,
1651
+ "rewards/margins": 0.28546684980392456,
1652
+ "rewards/margins_max": 0.5176615715026855,
1653
+ "rewards/margins_min": 0.09277740865945816,
1654
+ "rewards/margins_std": 0.1960216760635376,
1655
+ "rewards/rejected": -0.07862778007984161,
1656
+ "step": 740
1657
+ },
1658
+ {
1659
+ "dpo_losses": 0.5690917372703552,
1660
+ "epoch": 2.11,
1661
+ "grad_norm": 1.8282721044580463,
1662
+ "learning_rate": 1.219360813381446e-07,
1663
+ "logits/chosen": -2.6811976432800293,
1664
+ "logits/rejected": -2.6621005535125732,
1665
+ "logps/chosen": -266.5814514160156,
1666
+ "logps/rejected": -242.2976837158203,
1667
+ "loss": 0.5786,
1668
+ "positive_losses": 0.02016754075884819,
1669
+ "rewards/accuracies": 0.9750000238418579,
1670
+ "rewards/chosen": 0.20154385268688202,
1671
+ "rewards/margins": 0.27638477087020874,
1672
+ "rewards/margins_max": 0.4697951376438141,
1673
+ "rewards/margins_min": 0.1011359691619873,
1674
+ "rewards/margins_std": 0.16673585772514343,
1675
+ "rewards/rejected": -0.0748409777879715,
1676
+ "step": 750
1677
+ },
1678
+ {
1679
+ "dpo_losses": 0.5634604096412659,
1680
+ "epoch": 2.14,
1681
+ "grad_norm": 8.095290107453106,
1682
+ "learning_rate": 1.149652170686039e-07,
1683
+ "logits/chosen": -2.648603916168213,
1684
+ "logits/rejected": -2.6005616188049316,
1685
+ "logps/chosen": -267.0931396484375,
1686
+ "logps/rejected": -269.1485595703125,
1687
+ "loss": 0.5637,
1688
+ "positive_losses": 0.0,
1689
+ "rewards/accuracies": 0.9750000238418579,
1690
+ "rewards/chosen": 0.2069512903690338,
1691
+ "rewards/margins": 0.2928617596626282,
1692
+ "rewards/margins_max": 0.559156596660614,
1693
+ "rewards/margins_min": 0.08480425924062729,
1694
+ "rewards/margins_std": 0.21423542499542236,
1695
+ "rewards/rejected": -0.08591042459011078,
1696
+ "step": 760
1697
+ },
1698
+ {
1699
+ "dpo_losses": 0.5496206879615784,
1700
+ "epoch": 2.17,
1701
+ "grad_norm": 6.6085926070789,
1702
+ "learning_rate": 1.0813955575503587e-07,
1703
+ "logits/chosen": -2.6958391666412354,
1704
+ "logits/rejected": -2.6329185962677,
1705
+ "logps/chosen": -281.89190673828125,
1706
+ "logps/rejected": -222.1160125732422,
1707
+ "loss": 0.5739,
1708
+ "positive_losses": 0.15439815819263458,
1709
+ "rewards/accuracies": 1.0,
1710
+ "rewards/chosen": 0.22945666313171387,
1711
+ "rewards/margins": 0.3267485499382019,
1712
+ "rewards/margins_max": 0.5543676018714905,
1713
+ "rewards/margins_min": 0.1200588196516037,
1714
+ "rewards/margins_std": 0.20475919544696808,
1715
+ "rewards/rejected": -0.09729186445474625,
1716
+ "step": 770
1717
+ },
1718
+ {
1719
+ "dpo_losses": 0.5734694004058838,
1720
+ "epoch": 2.2,
1721
+ "grad_norm": 5.752056251728322,
1722
+ "learning_rate": 1.0146643703377486e-07,
1723
+ "logits/chosen": -2.751596212387085,
1724
+ "logits/rejected": -2.658079147338867,
1725
+ "logps/chosen": -297.37677001953125,
1726
+ "logps/rejected": -233.66806030273438,
1727
+ "loss": 0.6086,
1728
+ "positive_losses": 0.2737409174442291,
1729
+ "rewards/accuracies": 0.949999988079071,
1730
+ "rewards/chosen": 0.20195269584655762,
1731
+ "rewards/margins": 0.268646776676178,
1732
+ "rewards/margins_max": 0.4993320405483246,
1733
+ "rewards/margins_min": 0.051596127450466156,
1734
+ "rewards/margins_std": 0.20203988254070282,
1735
+ "rewards/rejected": -0.06669410318136215,
1736
+ "step": 780
1737
+ },
1738
+ {
1739
+ "dpo_losses": 0.5670967102050781,
1740
+ "epoch": 2.23,
1741
+ "grad_norm": 2.435776321138929,
1742
+ "learning_rate": 9.495303651204494e-08,
1743
+ "logits/chosen": -2.712763786315918,
1744
+ "logits/rejected": -2.630561113357544,
1745
+ "logps/chosen": -300.8298645019531,
1746
+ "logps/rejected": -252.09829711914062,
1747
+ "loss": 0.5701,
1748
+ "positive_losses": 0.03285813331604004,
1749
+ "rewards/accuracies": 0.9750000238418579,
1750
+ "rewards/chosen": 0.2152232676744461,
1751
+ "rewards/margins": 0.2839241027832031,
1752
+ "rewards/margins_max": 0.5225785970687866,
1753
+ "rewards/margins_min": 0.08764372766017914,
1754
+ "rewards/margins_std": 0.1984366476535797,
1755
+ "rewards/rejected": -0.06870082765817642,
1756
+ "step": 790
1757
+ },
1758
+ {
1759
+ "dpo_losses": 0.5538218021392822,
1760
+ "epoch": 2.25,
1761
+ "grad_norm": 7.082501471048372,
1762
+ "learning_rate": 8.860635805202615e-08,
1763
+ "logits/chosen": -2.7329680919647217,
1764
+ "logits/rejected": -2.6982383728027344,
1765
+ "logps/chosen": -299.06622314453125,
1766
+ "logps/rejected": -320.72845458984375,
1767
+ "loss": 0.5636,
1768
+ "positive_losses": 0.2872983515262604,
1769
+ "rewards/accuracies": 0.987500011920929,
1770
+ "rewards/chosen": 0.2248470038175583,
1771
+ "rewards/margins": 0.3167414963245392,
1772
+ "rewards/margins_max": 0.5594197511672974,
1773
+ "rewards/margins_min": 0.11445317417383194,
1774
+ "rewards/margins_std": 0.19880005717277527,
1775
+ "rewards/rejected": -0.0918944776058197,
1776
+ "step": 800
1777
+ },
1778
+ {
1779
+ "epoch": 2.25,
1780
+ "eval_dpo_losses": 0.6666676998138428,
1781
+ "eval_logits/chosen": -2.7115840911865234,
1782
+ "eval_logits/rejected": -2.6734206676483154,
1783
+ "eval_logps/chosen": -278.4820251464844,
1784
+ "eval_logps/rejected": -258.64727783203125,
1785
+ "eval_loss": 0.9370628595352173,
1786
+ "eval_positive_losses": 2.4201414585113525,
1787
+ "eval_rewards/accuracies": 0.6259999871253967,
1788
+ "eval_rewards/chosen": 0.06111405789852142,
1789
+ "eval_rewards/margins": 0.06179738789796829,
1790
+ "eval_rewards/margins_max": 0.3369947075843811,
1791
+ "eval_rewards/margins_min": -0.17766639590263367,
1792
+ "eval_rewards/margins_std": 0.17163515090942383,
1793
+ "eval_rewards/rejected": -0.0006833283696323633,
1794
+ "eval_runtime": 428.0428,
1795
+ "eval_samples_per_second": 4.672,
1796
+ "eval_steps_per_second": 0.292,
1797
+ "step": 800
1798
+ },
1799
+ {
1800
+ "dpo_losses": 0.561680793762207,
1801
+ "epoch": 2.28,
1802
+ "grad_norm": 6.853436586712954,
1803
+ "learning_rate": 8.24332262395994e-08,
1804
+ "logits/chosen": -2.770097255706787,
1805
+ "logits/rejected": -2.7309253215789795,
1806
+ "logps/chosen": -273.7652587890625,
1807
+ "logps/rejected": -278.20025634765625,
1808
+ "loss": 0.5736,
1809
+ "positive_losses": 0.0,
1810
+ "rewards/accuracies": 0.987500011920929,
1811
+ "rewards/chosen": 0.21009182929992676,
1812
+ "rewards/margins": 0.2937797009944916,
1813
+ "rewards/margins_max": 0.5182637572288513,
1814
+ "rewards/margins_min": 0.0909772664308548,
1815
+ "rewards/margins_std": 0.1924486607313156,
1816
+ "rewards/rejected": -0.0836879163980484,
1817
+ "step": 810
1818
+ },
1819
+ {
1820
+ "dpo_losses": 0.5531629920005798,
1821
+ "epoch": 2.31,
1822
+ "grad_norm": 2.2576986633959164,
1823
+ "learning_rate": 7.644027904586586e-08,
1824
+ "logits/chosen": -2.707874059677124,
1825
+ "logits/rejected": -2.6698241233825684,
1826
+ "logps/chosen": -305.3839416503906,
1827
+ "logps/rejected": -288.03485107421875,
1828
+ "loss": 0.5736,
1829
+ "positive_losses": 0.12947359681129456,
1830
+ "rewards/accuracies": 0.9624999761581421,
1831
+ "rewards/chosen": 0.24521002173423767,
1832
+ "rewards/margins": 0.32028818130493164,
1833
+ "rewards/margins_max": 0.5808194875717163,
1834
+ "rewards/margins_min": 0.09252934157848358,
1835
+ "rewards/margins_std": 0.21892566978931427,
1836
+ "rewards/rejected": -0.07507814466953278,
1837
+ "step": 820
1838
+ },
1839
+ {
1840
+ "dpo_losses": 0.5370634198188782,
1841
+ "epoch": 2.34,
1842
+ "grad_norm": 2.1285258942046736,
1843
+ "learning_rate": 7.063396068933469e-08,
1844
+ "logits/chosen": -2.7070274353027344,
1845
+ "logits/rejected": -2.6111221313476562,
1846
+ "logps/chosen": -358.33660888671875,
1847
+ "logps/rejected": -278.06463623046875,
1848
+ "loss": 0.5593,
1849
+ "positive_losses": 0.08015155792236328,
1850
+ "rewards/accuracies": 0.9624999761581421,
1851
+ "rewards/chosen": 0.2682201564311981,
1852
+ "rewards/margins": 0.35513558983802795,
1853
+ "rewards/margins_max": 0.5836988091468811,
1854
+ "rewards/margins_min": 0.14265316724777222,
1855
+ "rewards/margins_std": 0.19289033114910126,
1856
+ "rewards/rejected": -0.08691541850566864,
1857
+ "step": 830
1858
+ },
1859
+ {
1860
+ "dpo_losses": 0.5690010786056519,
1861
+ "epoch": 2.37,
1862
+ "grad_norm": 8.70012334977324,
1863
+ "learning_rate": 6.502051470645148e-08,
1864
+ "logits/chosen": -2.7239556312561035,
1865
+ "logits/rejected": -2.6661264896392822,
1866
+ "logps/chosen": -305.3841857910156,
1867
+ "logps/rejected": -253.9221649169922,
1868
+ "loss": 0.5836,
1869
+ "positive_losses": 0.07451782375574112,
1870
+ "rewards/accuracies": 0.949999988079071,
1871
+ "rewards/chosen": 0.20678095519542694,
1872
+ "rewards/margins": 0.2766776978969574,
1873
+ "rewards/margins_max": 0.5134538412094116,
1874
+ "rewards/margins_min": 0.10238063335418701,
1875
+ "rewards/margins_std": 0.18080976605415344,
1876
+ "rewards/rejected": -0.06989672034978867,
1877
+ "step": 840
1878
+ },
1879
+ {
1880
+ "dpo_losses": 0.5521928071975708,
1881
+ "epoch": 2.39,
1882
+ "grad_norm": 2.0434453054659096,
1883
+ "learning_rate": 5.960597723792194e-08,
1884
+ "logits/chosen": -2.7380244731903076,
1885
+ "logits/rejected": -2.6666131019592285,
1886
+ "logps/chosen": -271.77508544921875,
1887
+ "logps/rejected": -247.48190307617188,
1888
+ "loss": 0.5717,
1889
+ "positive_losses": 0.22315779328346252,
1890
+ "rewards/accuracies": 0.9624999761581421,
1891
+ "rewards/chosen": 0.2434767186641693,
1892
+ "rewards/margins": 0.3210228681564331,
1893
+ "rewards/margins_max": 0.5941131711006165,
1894
+ "rewards/margins_min": 0.11233635246753693,
1895
+ "rewards/margins_std": 0.21944975852966309,
1896
+ "rewards/rejected": -0.077546127140522,
1897
+ "step": 850
1898
+ },
1899
+ {
1900
+ "dpo_losses": 0.5401885509490967,
1901
+ "epoch": 2.42,
1902
+ "grad_norm": 2.1235783253077907,
1903
+ "learning_rate": 5.4396170538046486e-08,
1904
+ "logits/chosen": -2.7592415809631348,
1905
+ "logits/rejected": -2.7021079063415527,
1906
+ "logps/chosen": -321.6221923828125,
1907
+ "logps/rejected": -266.2222900390625,
1908
+ "loss": 0.5618,
1909
+ "positive_losses": 0.12075729668140411,
1910
+ "rewards/accuracies": 0.9750000238418579,
1911
+ "rewards/chosen": 0.2578734755516052,
1912
+ "rewards/margins": 0.35080546140670776,
1913
+ "rewards/margins_max": 0.625725269317627,
1914
+ "rewards/margins_min": 0.10036028921604156,
1915
+ "rewards/margins_std": 0.2362469732761383,
1916
+ "rewards/rejected": -0.09293195605278015,
1917
+ "step": 860
1918
+ },
1919
+ {
1920
+ "dpo_losses": 0.5563656687736511,
1921
+ "epoch": 2.45,
1922
+ "grad_norm": 1.8366012996504069,
1923
+ "learning_rate": 4.93966967140487e-08,
1924
+ "logits/chosen": -2.7032082080841064,
1925
+ "logits/rejected": -2.6532039642333984,
1926
+ "logps/chosen": -290.3298034667969,
1927
+ "logps/rejected": -287.1317138671875,
1928
+ "loss": 0.5634,
1929
+ "positive_losses": 0.1806434690952301,
1930
+ "rewards/accuracies": 0.9624999761581421,
1931
+ "rewards/chosen": 0.2245173156261444,
1932
+ "rewards/margins": 0.3118094205856323,
1933
+ "rewards/margins_max": 0.5898565053939819,
1934
+ "rewards/margins_min": 0.10215729475021362,
1935
+ "rewards/margins_std": 0.22476975619792938,
1936
+ "rewards/rejected": -0.08729207515716553,
1937
+ "step": 870
1938
+ },
1939
+ {
1940
+ "dpo_losses": 0.5541495084762573,
1941
+ "epoch": 2.48,
1942
+ "grad_norm": 6.44345291592453,
1943
+ "learning_rate": 4.4612931702126433e-08,
1944
+ "logits/chosen": -2.8107595443725586,
1945
+ "logits/rejected": -2.7267684936523438,
1946
+ "logps/chosen": -304.7294006347656,
1947
+ "logps/rejected": -270.358642578125,
1948
+ "loss": 0.5863,
1949
+ "positive_losses": 0.24675989151000977,
1950
+ "rewards/accuracies": 1.0,
1951
+ "rewards/chosen": 0.2371862679719925,
1952
+ "rewards/margins": 0.31598663330078125,
1953
+ "rewards/margins_max": 0.5958508253097534,
1954
+ "rewards/margins_min": 0.08327536284923553,
1955
+ "rewards/margins_std": 0.23089298605918884,
1956
+ "rewards/rejected": -0.07880039513111115,
1957
+ "step": 880
1958
+ },
1959
+ {
1960
+ "dpo_losses": 0.5567336082458496,
1961
+ "epoch": 2.51,
1962
+ "grad_norm": 7.843101687444187,
1963
+ "learning_rate": 4.005001948670605e-08,
1964
+ "logits/chosen": -2.775538682937622,
1965
+ "logits/rejected": -2.697503089904785,
1966
+ "logps/chosen": -322.6019287109375,
1967
+ "logps/rejected": -273.635986328125,
1968
+ "loss": 0.5736,
1969
+ "positive_losses": 0.37784984707832336,
1970
+ "rewards/accuracies": 0.9624999761581421,
1971
+ "rewards/chosen": 0.2152644395828247,
1972
+ "rewards/margins": 0.3087191581726074,
1973
+ "rewards/margins_max": 0.5784704685211182,
1974
+ "rewards/margins_min": 0.09841950982809067,
1975
+ "rewards/margins_std": 0.21595220267772675,
1976
+ "rewards/rejected": -0.09345470368862152,
1977
+ "step": 890
1978
+ },
1979
+ {
1980
+ "dpo_losses": 0.543450117111206,
1981
+ "epoch": 2.54,
1982
+ "grad_norm": 8.016670976771136,
1983
+ "learning_rate": 3.571286656911376e-08,
1984
+ "logits/chosen": -2.77311635017395,
1985
+ "logits/rejected": -2.6821038722991943,
1986
+ "logps/chosen": -304.2124938964844,
1987
+ "logps/rejected": -293.931884765625,
1988
+ "loss": 0.5736,
1989
+ "positive_losses": 0.34936437010765076,
1990
+ "rewards/accuracies": 0.9624999761581421,
1991
+ "rewards/chosen": 0.2413357049226761,
1992
+ "rewards/margins": 0.34193912148475647,
1993
+ "rewards/margins_max": 0.5774747133255005,
1994
+ "rewards/margins_min": 0.10269900411367416,
1995
+ "rewards/margins_std": 0.21173134446144104,
1996
+ "rewards/rejected": -0.10060342401266098,
1997
+ "step": 900
1998
+ },
1999
+ {
2000
+ "epoch": 2.54,
2001
+ "eval_dpo_losses": 0.6658776998519897,
2002
+ "eval_logits/chosen": -2.7107093334198,
2003
+ "eval_logits/rejected": -2.672572612762451,
2004
+ "eval_logps/chosen": -278.8089904785156,
2005
+ "eval_logps/rejected": -259.1817321777344,
2006
+ "eval_loss": 0.9591056704521179,
2007
+ "eval_positive_losses": 2.626781702041626,
2008
+ "eval_rewards/accuracies": 0.6320000290870667,
2009
+ "eval_rewards/chosen": 0.0578441321849823,
2010
+ "eval_rewards/margins": 0.06387220323085785,
2011
+ "eval_rewards/margins_max": 0.34665346145629883,
2012
+ "eval_rewards/margins_min": -0.18229486048221588,
2013
+ "eval_rewards/margins_std": 0.1764223277568817,
2014
+ "eval_rewards/rejected": -0.006028064992278814,
2015
+ "eval_runtime": 427.6906,
2016
+ "eval_samples_per_second": 4.676,
2017
+ "eval_steps_per_second": 0.292,
2018
+ "step": 900
2019
+ },
2020
+ {
2021
+ "dpo_losses": 0.5519391894340515,
2022
+ "epoch": 2.56,
2023
+ "grad_norm": 25.625793147024005,
2024
+ "learning_rate": 3.160613669161255e-08,
2025
+ "logits/chosen": -2.73115873336792,
2026
+ "logits/rejected": -2.670806407928467,
2027
+ "logps/chosen": -299.96832275390625,
2028
+ "logps/rejected": -239.826904296875,
2029
+ "loss": 0.5768,
2030
+ "positive_losses": 0.3415166437625885,
2031
+ "rewards/accuracies": 0.9375,
2032
+ "rewards/chosen": 0.23844274878501892,
2033
+ "rewards/margins": 0.3223797678947449,
2034
+ "rewards/margins_max": 0.5813696980476379,
2035
+ "rewards/margins_min": 0.09977176040410995,
2036
+ "rewards/margins_std": 0.21656760573387146,
2037
+ "rewards/rejected": -0.08393705636262894,
2038
+ "step": 910
2039
+ },
2040
+ {
2041
+ "dpo_losses": 0.5531378388404846,
2042
+ "epoch": 2.59,
2043
+ "grad_norm": 2.1656044654918523,
2044
+ "learning_rate": 2.7734245822478436e-08,
2045
+ "logits/chosen": -2.7032344341278076,
2046
+ "logits/rejected": -2.612525463104248,
2047
+ "logps/chosen": -289.3112487792969,
2048
+ "logps/rejected": -218.96713256835938,
2049
+ "loss": 0.5749,
2050
+ "positive_losses": 0.18140992522239685,
2051
+ "rewards/accuracies": 0.9624999761581421,
2052
+ "rewards/chosen": 0.2416727989912033,
2053
+ "rewards/margins": 0.32405149936676025,
2054
+ "rewards/margins_max": 0.6709355115890503,
2055
+ "rewards/margins_min": 0.08473904430866241,
2056
+ "rewards/margins_std": 0.26692530512809753,
2057
+ "rewards/rejected": -0.08237870782613754,
2058
+ "step": 920
2059
+ },
2060
+ {
2061
+ "dpo_losses": 0.5606621503829956,
2062
+ "epoch": 2.62,
2063
+ "grad_norm": 8.473170854444827,
2064
+ "learning_rate": 2.410135740750821e-08,
2065
+ "logits/chosen": -2.735752582550049,
2066
+ "logits/rejected": -2.671776294708252,
2067
+ "logps/chosen": -259.04132080078125,
2068
+ "logps/rejected": -235.8688201904297,
2069
+ "loss": 0.5834,
2070
+ "positive_losses": 0.36757582426071167,
2071
+ "rewards/accuracies": 0.987500011920929,
2072
+ "rewards/chosen": 0.22701683640480042,
2073
+ "rewards/margins": 0.30118703842163086,
2074
+ "rewards/margins_max": 0.5452762246131897,
2075
+ "rewards/margins_min": 0.08702322095632553,
2076
+ "rewards/margins_std": 0.20802298188209534,
2077
+ "rewards/rejected": -0.07417017966508865,
2078
+ "step": 930
2079
+ },
2080
+ {
2081
+ "dpo_losses": 0.5595995187759399,
2082
+ "epoch": 2.65,
2083
+ "grad_norm": 6.88019656676779,
2084
+ "learning_rate": 2.071137789306418e-08,
2085
+ "logits/chosen": -2.7638261318206787,
2086
+ "logits/rejected": -2.685157537460327,
2087
+ "logps/chosen": -326.85858154296875,
2088
+ "logps/rejected": -302.33209228515625,
2089
+ "loss": 0.572,
2090
+ "positive_losses": 0.20179709792137146,
2091
+ "rewards/accuracies": 0.949999988079071,
2092
+ "rewards/chosen": 0.2141149938106537,
2093
+ "rewards/margins": 0.3031490445137024,
2094
+ "rewards/margins_max": 0.5786975026130676,
2095
+ "rewards/margins_min": 0.07930903881788254,
2096
+ "rewards/margins_std": 0.2305254191160202,
2097
+ "rewards/rejected": -0.08903402090072632,
2098
+ "step": 940
2099
+ },
2100
+ {
2101
+ "dpo_losses": 0.5905728936195374,
2102
+ "epoch": 2.68,
2103
+ "grad_norm": 2.0169170880957616,
2104
+ "learning_rate": 1.7567952525471107e-08,
2105
+ "logits/chosen": -2.696136951446533,
2106
+ "logits/rejected": -2.6454825401306152,
2107
+ "logps/chosen": -245.9570770263672,
2108
+ "logps/rejected": -213.52334594726562,
2109
+ "loss": 0.6002,
2110
+ "positive_losses": 0.25799694657325745,
2111
+ "rewards/accuracies": 0.9624999761581421,
2112
+ "rewards/chosen": 0.1696554720401764,
2113
+ "rewards/margins": 0.22757229208946228,
2114
+ "rewards/margins_max": 0.4704417288303375,
2115
+ "rewards/margins_min": 0.05507839843630791,
2116
+ "rewards/margins_std": 0.18926933407783508,
2117
+ "rewards/rejected": -0.057916827499866486,
2118
+ "step": 950
2119
+ },
2120
+ {
2121
+ "dpo_losses": 0.560580313205719,
2122
+ "epoch": 2.7,
2123
+ "grad_norm": 7.04290537877064,
2124
+ "learning_rate": 1.467446143128101e-08,
2125
+ "logits/chosen": -2.787949800491333,
2126
+ "logits/rejected": -2.7239291667938232,
2127
+ "logps/chosen": -260.6053161621094,
2128
+ "logps/rejected": -223.7249755859375,
2129
+ "loss": 0.5671,
2130
+ "positive_losses": 0.06990127265453339,
2131
+ "rewards/accuracies": 0.9750000238418579,
2132
+ "rewards/chosen": 0.22634287178516388,
2133
+ "rewards/margins": 0.29788738489151,
2134
+ "rewards/margins_max": 0.5511468648910522,
2135
+ "rewards/margins_min": 0.08614132553339005,
2136
+ "rewards/margins_std": 0.2106626331806183,
2137
+ "rewards/rejected": -0.07154452800750732,
2138
+ "step": 960
2139
+ },
2140
+ {
2141
+ "dpo_losses": 0.5514933466911316,
2142
+ "epoch": 2.73,
2143
+ "grad_norm": 1.9051880301646325,
2144
+ "learning_rate": 1.2034015982622243e-08,
2145
+ "logits/chosen": -2.7561697959899902,
2146
+ "logits/rejected": -2.6536028385162354,
2147
+ "logps/chosen": -329.5728759765625,
2148
+ "logps/rejected": -290.1389465332031,
2149
+ "loss": 0.5909,
2150
+ "positive_losses": 0.08531048148870468,
2151
+ "rewards/accuracies": 1.0,
2152
+ "rewards/chosen": 0.22805485129356384,
2153
+ "rewards/margins": 0.3209593594074249,
2154
+ "rewards/margins_max": 0.5957485437393188,
2155
+ "rewards/margins_min": 0.09393687546253204,
2156
+ "rewards/margins_std": 0.2260352075099945,
2157
+ "rewards/rejected": -0.09290449321269989,
2158
+ "step": 970
2159
+ },
2160
+ {
2161
+ "dpo_losses": 0.5581797957420349,
2162
+ "epoch": 2.76,
2163
+ "grad_norm": 13.216663962544713,
2164
+ "learning_rate": 9.649455451539418e-09,
2165
+ "logits/chosen": -2.6970033645629883,
2166
+ "logits/rejected": -2.651496410369873,
2167
+ "logps/chosen": -266.47296142578125,
2168
+ "logps/rejected": -233.91098022460938,
2169
+ "loss": 0.5837,
2170
+ "positive_losses": 0.05353069305419922,
2171
+ "rewards/accuracies": 1.0,
2172
+ "rewards/chosen": 0.21983060240745544,
2173
+ "rewards/margins": 0.30383530259132385,
2174
+ "rewards/margins_max": 0.517738401889801,
2175
+ "rewards/margins_min": 0.09517470002174377,
2176
+ "rewards/margins_std": 0.19478817284107208,
2177
+ "rewards/rejected": -0.08400467783212662,
2178
+ "step": 980
2179
+ },
2180
+ {
2181
+ "dpo_losses": 0.5490652322769165,
2182
+ "epoch": 2.79,
2183
+ "grad_norm": 2.0486277290431736,
2184
+ "learning_rate": 7.523343956923194e-09,
2185
+ "logits/chosen": -2.7556817531585693,
2186
+ "logits/rejected": -2.6662416458129883,
2187
+ "logps/chosen": -294.7388916015625,
2188
+ "logps/rejected": -267.9339904785156,
2189
+ "loss": 0.5647,
2190
+ "positive_losses": 0.10996627807617188,
2191
+ "rewards/accuracies": 0.9750000238418579,
2192
+ "rewards/chosen": 0.2385723888874054,
2193
+ "rewards/margins": 0.3286248743534088,
2194
+ "rewards/margins_max": 0.6235631108283997,
2195
+ "rewards/margins_min": 0.11854185163974762,
2196
+ "rewards/margins_std": 0.2294415682554245,
2197
+ "rewards/rejected": -0.09005247056484222,
2198
+ "step": 990
2199
+ },
2200
+ {
2201
+ "dpo_losses": 0.5738095045089722,
2202
+ "epoch": 2.82,
2203
+ "grad_norm": 6.03858746514603,
2204
+ "learning_rate": 5.6579677073121945e-09,
2205
+ "logits/chosen": -2.7139980792999268,
2206
+ "logits/rejected": -2.6967344284057617,
2207
+ "logps/chosen": -273.75885009765625,
2208
+ "logps/rejected": -276.884033203125,
2209
+ "loss": 0.5825,
2210
+ "positive_losses": 0.2053932249546051,
2211
+ "rewards/accuracies": 0.9750000238418579,
2212
+ "rewards/chosen": 0.1967349648475647,
2213
+ "rewards/margins": 0.2664409279823303,
2214
+ "rewards/margins_max": 0.49395784735679626,
2215
+ "rewards/margins_min": 0.09026306867599487,
2216
+ "rewards/margins_std": 0.1819494068622589,
2217
+ "rewards/rejected": -0.06970598548650742,
2218
+ "step": 1000
2219
+ },
2220
+ {
2221
+ "epoch": 2.82,
2222
+ "eval_dpo_losses": 0.665816068649292,
2223
+ "eval_logits/chosen": -2.712677001953125,
2224
+ "eval_logits/rejected": -2.674919366836548,
2225
+ "eval_logps/chosen": -278.6134033203125,
2226
+ "eval_logps/rejected": -259.0028076171875,
2227
+ "eval_loss": 0.9543355107307434,
2228
+ "eval_positive_losses": 2.581048011779785,
2229
+ "eval_rewards/accuracies": 0.6290000081062317,
2230
+ "eval_rewards/chosen": 0.0598001554608345,
2231
+ "eval_rewards/margins": 0.06403880566358566,
2232
+ "eval_rewards/margins_max": 0.34747231006622314,
2233
+ "eval_rewards/margins_min": -0.1826418787240982,
2234
+ "eval_rewards/margins_std": 0.17671293020248413,
2235
+ "eval_rewards/rejected": -0.004238648805767298,
2236
+ "eval_runtime": 427.7348,
2237
+ "eval_samples_per_second": 4.676,
2238
+ "eval_steps_per_second": 0.292,
2239
+ "step": 1000
2240
+ },
2241
+ {
2242
+ "dpo_losses": 0.5594234466552734,
2243
+ "epoch": 2.85,
2244
+ "grad_norm": 10.058501971668669,
2245
+ "learning_rate": 4.0553325425319585e-09,
2246
+ "logits/chosen": -2.789537191390991,
2247
+ "logits/rejected": -2.7190775871276855,
2248
+ "logps/chosen": -301.9007568359375,
2249
+ "logps/rejected": -313.83831787109375,
2250
+ "loss": 0.6005,
2251
+ "positive_losses": 0.10010738670825958,
2252
+ "rewards/accuracies": 0.987500011920929,
2253
+ "rewards/chosen": 0.21791288256645203,
2254
+ "rewards/margins": 0.30005866289138794,
2255
+ "rewards/margins_max": 0.5106627345085144,
2256
+ "rewards/margins_min": 0.13350318372249603,
2257
+ "rewards/margins_std": 0.18063752353191376,
2258
+ "rewards/rejected": -0.0821457952260971,
2259
+ "step": 1010
2260
+ },
2261
+ {
2262
+ "dpo_losses": 0.5577588081359863,
2263
+ "epoch": 2.87,
2264
+ "grad_norm": 1.8795598215831135,
2265
+ "learning_rate": 2.717161776814747e-09,
2266
+ "logits/chosen": -2.7825233936309814,
2267
+ "logits/rejected": -2.7229418754577637,
2268
+ "logps/chosen": -291.11932373046875,
2269
+ "logps/rejected": -272.02923583984375,
2270
+ "loss": 0.5722,
2271
+ "positive_losses": 0.11093978583812714,
2272
+ "rewards/accuracies": 0.9750000238418579,
2273
+ "rewards/chosen": 0.22510568797588348,
2274
+ "rewards/margins": 0.30276286602020264,
2275
+ "rewards/margins_max": 0.5086154341697693,
2276
+ "rewards/margins_min": 0.1081671267747879,
2277
+ "rewards/margins_std": 0.1743108481168747,
2278
+ "rewards/rejected": -0.07765716314315796,
2279
+ "step": 1020
2280
+ },
2281
+ {
2282
+ "dpo_losses": 0.555158257484436,
2283
+ "epoch": 2.9,
2284
+ "grad_norm": 1.7202517104603963,
2285
+ "learning_rate": 1.6448943457189613e-09,
2286
+ "logits/chosen": -2.6988253593444824,
2287
+ "logits/rejected": -2.6433205604553223,
2288
+ "logps/chosen": -286.4718322753906,
2289
+ "logps/rejected": -265.35052490234375,
2290
+ "loss": 0.5888,
2291
+ "positive_losses": 0.37335652112960815,
2292
+ "rewards/accuracies": 0.9624999761581421,
2293
+ "rewards/chosen": 0.23134306073188782,
2294
+ "rewards/margins": 0.31557995080947876,
2295
+ "rewards/margins_max": 0.594192624092102,
2296
+ "rewards/margins_min": 0.10240230709314346,
2297
+ "rewards/margins_std": 0.2274932563304901,
2298
+ "rewards/rejected": -0.08423684537410736,
2299
+ "step": 1030
2300
+ },
2301
+ {
2302
+ "dpo_losses": 0.5689266324043274,
2303
+ "epoch": 2.93,
2304
+ "grad_norm": 15.663649464629284,
2305
+ "learning_rate": 8.396832588411229e-10,
2306
+ "logits/chosen": -2.6604883670806885,
2307
+ "logits/rejected": -2.586012840270996,
2308
+ "logps/chosen": -273.3360290527344,
2309
+ "logps/rejected": -243.2672882080078,
2310
+ "loss": 0.5953,
2311
+ "positive_losses": 0.21942074596881866,
2312
+ "rewards/accuracies": 0.9750000238418579,
2313
+ "rewards/chosen": 0.21510818600654602,
2314
+ "rewards/margins": 0.2806702256202698,
2315
+ "rewards/margins_max": 0.5532082319259644,
2316
+ "rewards/margins_min": 0.06844080239534378,
2317
+ "rewards/margins_std": 0.22315998375415802,
2318
+ "rewards/rejected": -0.06556206941604614,
2319
+ "step": 1040
2320
+ },
2321
+ {
2322
+ "dpo_losses": 0.5607948303222656,
2323
+ "epoch": 2.96,
2324
+ "grad_norm": 7.720480782752179,
2325
+ "learning_rate": 3.0239435998430374e-10,
2326
+ "logits/chosen": -2.7420949935913086,
2327
+ "logits/rejected": -2.663933038711548,
2328
+ "logps/chosen": -278.32171630859375,
2329
+ "logps/rejected": -232.65087890625,
2330
+ "loss": 0.58,
2331
+ "positive_losses": 0.3233460783958435,
2332
+ "rewards/accuracies": 0.9375,
2333
+ "rewards/chosen": 0.23241767287254333,
2334
+ "rewards/margins": 0.2982451915740967,
2335
+ "rewards/margins_max": 0.537132978439331,
2336
+ "rewards/margins_min": 0.06783739477396011,
2337
+ "rewards/margins_std": 0.2153671681880951,
2338
+ "rewards/rejected": -0.06582748889923096,
2339
+ "step": 1050
2340
+ },
2341
+ {
2342
+ "dpo_losses": 0.5575666427612305,
2343
+ "epoch": 2.99,
2344
+ "grad_norm": 1.9829415978985032,
2345
+ "learning_rate": 3.360539611582669e-11,
2346
+ "logits/chosen": -2.6903703212738037,
2347
+ "logits/rejected": -2.675687789916992,
2348
+ "logps/chosen": -256.32183837890625,
2349
+ "logps/rejected": -261.22113037109375,
2350
+ "loss": 0.5684,
2351
+ "positive_losses": 0.07837722450494766,
2352
+ "rewards/accuracies": 0.9750000238418579,
2353
+ "rewards/chosen": 0.23069465160369873,
2354
+ "rewards/margins": 0.30815768241882324,
2355
+ "rewards/margins_max": 0.5907589793205261,
2356
+ "rewards/margins_min": 0.066887766122818,
2357
+ "rewards/margins_std": 0.2346896380186081,
2358
+ "rewards/rejected": -0.0774630457162857,
2359
+ "step": 1060
2360
+ },
2361
+ {
2362
+ "epoch": 3.0,
2363
+ "step": 1065,
2364
+ "total_flos": 0.0,
2365
+ "train_loss": 0.6146632369135467,
2366
+ "train_runtime": 13546.83,
2367
+ "train_samples_per_second": 1.257,
2368
+ "train_steps_per_second": 0.079
2369
+ }
2370
+ ],
2371
+ "logging_steps": 10,
2372
+ "max_steps": 1065,
2373
+ "num_input_tokens_seen": 0,
2374
+ "num_train_epochs": 3,
2375
+ "save_steps": 100,
2376
+ "total_flos": 0.0,
2377
+ "train_batch_size": 4,
2378
+ "trial_name": null,
2379
+ "trial_params": null
2380
+ }