hZzy commited on
Commit
dc3c59b
1 Parent(s): 6e1bd11

Model save

Browse files
Files changed (5) hide show
  1. README.md +22 -19
  2. all_results.json +4 -4
  3. model.safetensors +1 -1
  4. train_results.json +4 -4
  5. trainer_state.json +1875 -9
README.md CHANGED
@@ -2,18 +2,11 @@
2
  license: apache-2.0
3
  base_model: hZzy/qwen2.5-0.5b-sft-news-IFT
4
  tags:
5
- - alignment-handbook
6
- - ndcg
7
- - trl
8
- - expo
9
- - generated_from_trainer
10
  - trl
11
  - expo
12
  - alignment-handbook
13
  - ndcg
14
  - generated_from_trainer
15
- datasets:
16
- - hZzy/train_pairwise
17
  model-index:
18
  - name: qwen2.5-0.5b-expo-DPO-ES-TRY
19
  results: []
@@ -22,21 +15,21 @@ model-index:
22
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
23
  should probably proofread and complete it, then remove this comment. -->
24
 
25
- [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/zhiyuzha-university-of-florida/huggingface/runs/8qngt5rk)
26
  # qwen2.5-0.5b-expo-DPO-ES-TRY
27
 
28
- This model is a fine-tuned version of [hZzy/qwen2.5-0.5b-sft-news-IFT](https://huggingface.co/hZzy/qwen2.5-0.5b-sft-news-IFT) on the hZzy/train_pairwise dataset.
29
  It achieves the following results on the evaluation set:
30
- - Loss: 0.8842
31
- - Logps: -122.3753
32
- - Logits: -2.2914
33
- - Objective: 0.9181
34
- - Dpo Loss: 0.9181
35
- - Regularize: 0.9181
36
- - Ranking Simple: 0.5642
37
  - Ranking Idealized: 0.6046
38
  - Ranking Idealized Expo: 0.5280
39
- - Dpo Wo Beta: -6.4269
40
 
41
  ## Model description
42
 
@@ -61,8 +54,8 @@ The following hyperparameters were used during training:
61
  - seed: 42
62
  - distributed_type: multi-GPU
63
  - num_devices: 6
64
- - gradient_accumulation_steps: 12
65
- - total_train_batch_size: 144
66
  - total_eval_batch_size: 12
67
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
68
  - lr_scheduler_type: cosine
@@ -82,6 +75,16 @@ The following hyperparameters were used during training:
82
  | 0.1227 | 2.1030 | 371 | 0.9224 | -6.4510 | -1.8645 | -122.0016 | 0.8844 | 0.9224 | 0.5888 | 0.5093 | 0.5424 | 0.9224 |
83
  | 0.133 | 2.4034 | 424 | 0.8786 | -5.8878 | -2.0277 | -117.1217 | 0.8448 | 0.8786 | 0.5888 | 0.5093 | 0.5413 | 0.8786 |
84
  | 0.1211 | 2.7085 | 477 | 0.8739 | -5.8152 | -2.0272 | -116.4230 | 0.8371 | 0.8739 | 0.5888 | 0.5093 | 0.5403 | 0.8739 |
 
 
 
 
 
 
 
 
 
 
85
 
86
 
87
  ### Framework versions
 
2
  license: apache-2.0
3
  base_model: hZzy/qwen2.5-0.5b-sft-news-IFT
4
  tags:
 
 
 
 
 
5
  - trl
6
  - expo
7
  - alignment-handbook
8
  - ndcg
9
  - generated_from_trainer
 
 
10
  model-index:
11
  - name: qwen2.5-0.5b-expo-DPO-ES-TRY
12
  results: []
 
15
  <!-- This model card has been generated automatically according to the information the Trainer had access to. You
16
  should probably proofread and complete it, then remove this comment. -->
17
 
18
+ [<img src="https://raw.githubusercontent.com/wandb/assets/main/wandb-github-badge-28.svg" alt="Visualize in Weights & Biases" width="200" height="32"/>](https://wandb.ai/zhiyuzha-university-of-florida/huggingface/runs/rg2tu2w0)
19
  # qwen2.5-0.5b-expo-DPO-ES-TRY
20
 
21
+ This model is a fine-tuned version of [hZzy/qwen2.5-0.5b-sft-news-IFT](https://huggingface.co/hZzy/qwen2.5-0.5b-sft-news-IFT) on an unknown dataset.
22
  It achieves the following results on the evaluation set:
23
+ - Loss: 0.8682
24
+ - Logps: -123.1454
25
+ - Logits: -2.5127
26
+ - Objective: 0.8882
27
+ - Dpo Loss: 0.8882
28
+ - Regularize: 0.8882
29
+ - Ranking Simple: 0.5600
30
  - Ranking Idealized: 0.6046
31
  - Ranking Idealized Expo: 0.5280
32
+ - Dpo Wo Beta: -6.1537
33
 
34
  ## Model description
35
 
 
54
  - seed: 42
55
  - distributed_type: multi-GPU
56
  - num_devices: 6
57
+ - gradient_accumulation_steps: 6
58
+ - total_train_batch_size: 72
59
  - total_eval_batch_size: 12
60
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
61
  - lr_scheduler_type: cosine
 
75
  | 0.1227 | 2.1030 | 371 | 0.9224 | -6.4510 | -1.8645 | -122.0016 | 0.8844 | 0.9224 | 0.5888 | 0.5093 | 0.5424 | 0.9224 |
76
  | 0.133 | 2.4034 | 424 | 0.8786 | -5.8878 | -2.0277 | -117.1217 | 0.8448 | 0.8786 | 0.5888 | 0.5093 | 0.5413 | 0.8786 |
77
  | 0.1211 | 2.7085 | 477 | 0.8739 | -5.8152 | -2.0272 | -116.4230 | 0.8371 | 0.8739 | 0.5888 | 0.5093 | 0.5403 | 0.8739 |
78
+ | 0.0858 | 1.5045 | 530 | 0.8753 | -5.9229 | -2.4530 | -118.2529 | 0.8505 | 0.8753 | 0.6046 | 0.5280 | 0.5683 | 0.8753 |
79
+ | 0.1274 | 1.6547 | 583 | 0.8264 | -5.2847 | -2.4380 | -119.5907 | 0.8086 | 0.8264 | 0.6046 | 0.5280 | 0.5870 | 0.8264 |
80
+ | 0.1614 | 1.8049 | 636 | 0.8243 | -5.2813 | -2.4850 | -117.8585 | 0.8209 | 0.8243 | 0.6046 | 0.5280 | 0.5818 | 0.8243 |
81
+ | 0.1616 | 1.9551 | 689 | 0.8576 | -5.7234 | -2.4656 | -119.3221 | 0.8383 | 0.8576 | 0.6046 | 0.5280 | 0.5797 | 0.8576 |
82
+ | 0.1063 | 2.1053 | 742 | 0.9824 | -7.3310 | -2.2712 | -133.3637 | 0.9486 | 0.9824 | 0.6046 | 0.5280 | 0.5518 | 0.9824 |
83
+ | 0.1017 | 2.2556 | 795 | 0.8904 | -6.2055 | -2.4490 | -123.5745 | 0.8711 | 0.8904 | 0.6046 | 0.5280 | 0.5683 | 0.8904 |
84
+ | 0.1225 | 2.4058 | 848 | 0.9035 | -6.3529 | -2.4743 | -124.5336 | 0.8822 | 0.9035 | 0.6046 | 0.5280 | 0.5569 | 0.9035 |
85
+ | 0.1157 | 2.5583 | 901 | 0.8718 | -124.4583 | -2.4886 | 0.8941 | 0.8941 | 0.8941 | 0.5621 | 0.6046 | 0.5280 | -6.2136 |
86
+ | 0.1387 | 2.7085 | 954 | 0.8688 | -123.4143 | -2.5086 | 0.8892 | 0.8892 | 0.8892 | 0.5580 | 0.6046 | 0.5280 | -6.1610 |
87
+ | 0.1219 | 2.8588 | 1007 | 0.8682 | -123.1454 | -2.5127 | 0.8882 | 0.8882 | 0.8882 | 0.5600 | 0.6046 | 0.5280 | -6.1537 |
88
 
89
 
90
  ### Framework versions
all_results.json CHANGED
@@ -15,9 +15,9 @@
15
  "eval_samples_per_second": 18.1,
16
  "eval_steps_per_second": 1.51,
17
  "total_flos": 0.0,
18
- "train_loss": 0.0,
19
- "train_runtime": 9.3184,
20
  "train_samples": 50802,
21
- "train_samples_per_second": 16355.433,
22
- "train_steps_per_second": 56.662
23
  }
 
15
  "eval_samples_per_second": 18.1,
16
  "eval_steps_per_second": 1.51,
17
  "total_flos": 0.0,
18
+ "train_loss": 0.022545777056648425,
19
+ "train_runtime": 4386.5835,
20
  "train_samples": 50802,
21
+ "train_samples_per_second": 34.744,
22
+ "train_steps_per_second": 0.241
23
  }
model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:9f5bfe3c37fcb09994eb5ad4bf36742c81baf6cf4984d1788641eb270c7eda21
3
  size 1975192208
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:194207d30450b9e4443cc4f6e08ed7bac1e41c86af1d51b105835e6a19278e3d
3
  size 1975192208
train_results.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
  "epoch": 2.9976381672177608,
3
  "total_flos": 0.0,
4
- "train_loss": 0.0,
5
- "train_runtime": 9.3184,
6
  "train_samples": 50802,
7
- "train_samples_per_second": 16355.433,
8
- "train_steps_per_second": 56.662
9
  }
 
1
  {
2
  "epoch": 2.9976381672177608,
3
  "total_flos": 0.0,
4
+ "train_loss": 0.022545777056648425,
5
+ "train_runtime": 4386.5835,
6
  "train_samples": 50802,
7
+ "train_samples_per_second": 34.744,
8
+ "train_steps_per_second": 0.241
9
  }
trainer_state.json CHANGED
@@ -1,9 +1,9 @@
1
  {
2
- "best_metric": 0.5423553586006165,
3
- "best_model_checkpoint": "./qwen2.5-0.5b/qwen2.5-0.5b-expo-DPO-ES-TRY/checkpoint-371",
4
  "epoch": 2.9976381672177608,
5
  "eval_steps": 53,
6
- "global_step": 528,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
@@ -1857,18 +1857,1884 @@
1857
  "regularize": 0.13335375487804413,
1858
  "step": 525
1859
  },
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1860
  {
1861
  "epoch": 2.9976381672177608,
1862
- "step": 528,
1863
  "total_flos": 0.0,
1864
- "train_loss": 0.0,
1865
- "train_runtime": 9.3184,
1866
- "train_samples_per_second": 16355.433,
1867
- "train_steps_per_second": 56.662
1868
  }
1869
  ],
1870
  "logging_steps": 5,
1871
- "max_steps": 528,
1872
  "num_input_tokens_seen": 0,
1873
  "num_train_epochs": 3,
1874
  "save_steps": 53,
 
1
  {
2
+ "best_metric": 0.5869565010070801,
3
+ "best_model_checkpoint": "./qwen2.5-0.5b/qwen2.5-0.5b-expo-DPO-ES-TRY/checkpoint-583",
4
  "epoch": 2.9976381672177608,
5
  "eval_steps": 53,
6
+ "global_step": 1056,
7
  "is_hyper_param_search": false,
8
  "is_local_process_zero": true,
9
  "is_world_process_zero": true,
 
1857
  "regularize": 0.13335375487804413,
1858
  "step": 525
1859
  },
1860
+ {
1861
+ "dpo_loss": 0.08052093535661697,
1862
+ "dpo_wo_beta": -0.004379949066787958,
1863
+ "epoch": 1.5044874822862542,
1864
+ "grad_norm": 9.015683090612498,
1865
+ "learning_rate": 2.9196388040863695e-06,
1866
+ "logits": -1.833287239074707,
1867
+ "logps": -114.10733795166016,
1868
+ "loss": 0.0858,
1869
+ "objective": 0.08052093535661697,
1870
+ "ranking_idealized": 0.6666666865348816,
1871
+ "ranking_idealized_expo": 0.5416666865348816,
1872
+ "ranking_simple": 0.9375,
1873
+ "regularize": 0.08052093535661697,
1874
+ "step": 530
1875
+ },
1876
+ {
1877
+ "epoch": 1.5044874822862542,
1878
+ "eval_dpo_loss": 0.8753401041030884,
1879
+ "eval_dpo_wo_beta": -5.922874450683594,
1880
+ "eval_logits": -2.4529590606689453,
1881
+ "eval_logps": -118.25288391113281,
1882
+ "eval_loss": 0.8505071997642517,
1883
+ "eval_objective": 0.8753401041030884,
1884
+ "eval_ranking_idealized": 0.6045548915863037,
1885
+ "eval_ranking_idealized_expo": 0.5279502868652344,
1886
+ "eval_ranking_simple": 0.5683229565620422,
1887
+ "eval_regularize": 0.8753401041030884,
1888
+ "eval_runtime": 344.4459,
1889
+ "eval_samples_per_second": 16.81,
1890
+ "eval_steps_per_second": 1.402,
1891
+ "step": 530
1892
+ },
1893
+ {
1894
+ "dpo_loss": 0.06306228041648865,
1895
+ "dpo_wo_beta": -0.08629266172647476,
1896
+ "epoch": 1.5186584789796882,
1897
+ "grad_norm": 12.634106279395441,
1898
+ "learning_rate": 2.8788330549198512e-06,
1899
+ "logits": -1.8371531963348389,
1900
+ "logps": -115.510009765625,
1901
+ "loss": 0.0766,
1902
+ "objective": 0.06306228041648865,
1903
+ "ranking_idealized": 0.550000011920929,
1904
+ "ranking_idealized_expo": 0.4833333194255829,
1905
+ "ranking_simple": 0.9166666865348816,
1906
+ "regularize": 0.06306228041648865,
1907
+ "step": 535
1908
+ },
1909
+ {
1910
+ "dpo_loss": 0.13102607429027557,
1911
+ "dpo_wo_beta": -0.41934680938720703,
1912
+ "epoch": 1.5328294756731222,
1913
+ "grad_norm": 14.583212522491863,
1914
+ "learning_rate": 2.8379237365787426e-06,
1915
+ "logits": -1.7649121284484863,
1916
+ "logps": -127.44564819335938,
1917
+ "loss": 0.0928,
1918
+ "objective": 0.13102607429027557,
1919
+ "ranking_idealized": 0.5833333134651184,
1920
+ "ranking_idealized_expo": 0.5249999761581421,
1921
+ "ranking_simple": 0.8999999761581421,
1922
+ "regularize": 0.13102607429027557,
1923
+ "step": 540
1924
+ },
1925
+ {
1926
+ "dpo_loss": 0.12010473757982254,
1927
+ "dpo_wo_beta": -0.2892196476459503,
1928
+ "epoch": 1.5470004723665565,
1929
+ "grad_norm": 18.74636858022955,
1930
+ "learning_rate": 2.7969220332622004e-06,
1931
+ "logits": -1.7403244972229004,
1932
+ "logps": -126.89453125,
1933
+ "loss": 0.1114,
1934
+ "objective": 0.12010473757982254,
1935
+ "ranking_idealized": 0.6499999761581421,
1936
+ "ranking_idealized_expo": 0.550000011920929,
1937
+ "ranking_simple": 0.925000011920929,
1938
+ "regularize": 0.12010473757982254,
1939
+ "step": 545
1940
+ },
1941
+ {
1942
+ "dpo_loss": 0.05550822243094444,
1943
+ "dpo_wo_beta": -0.0006199590279720724,
1944
+ "epoch": 1.5611714690599907,
1945
+ "grad_norm": 16.65748534309405,
1946
+ "learning_rate": 2.7558391544265127e-06,
1947
+ "logits": -1.7434070110321045,
1948
+ "logps": -120.79520416259766,
1949
+ "loss": 0.0872,
1950
+ "objective": 0.05550822243094444,
1951
+ "ranking_idealized": 0.6000000238418579,
1952
+ "ranking_idealized_expo": 0.4833333194255829,
1953
+ "ranking_simple": 0.9750000238418579,
1954
+ "regularize": 0.05550822243094444,
1955
+ "step": 550
1956
+ },
1957
+ {
1958
+ "dpo_loss": 0.12240471690893173,
1959
+ "dpo_wo_beta": -0.22643856704235077,
1960
+ "epoch": 1.5753424657534247,
1961
+ "grad_norm": 12.27653222683965,
1962
+ "learning_rate": 2.714686331720543e-06,
1963
+ "logits": -1.8163702487945557,
1964
+ "logps": -123.55280303955078,
1965
+ "loss": 0.103,
1966
+ "objective": 0.12240471690893173,
1967
+ "ranking_idealized": 0.6083333492279053,
1968
+ "ranking_idealized_expo": 0.5416666865348816,
1969
+ "ranking_simple": 0.875,
1970
+ "regularize": 0.12240471690893173,
1971
+ "step": 555
1972
+ },
1973
+ {
1974
+ "dpo_loss": 0.1260891556739807,
1975
+ "dpo_wo_beta": -0.384922057390213,
1976
+ "epoch": 1.5895134624468588,
1977
+ "grad_norm": 14.054554594902122,
1978
+ "learning_rate": 2.6734748159151104e-06,
1979
+ "logits": -1.8347235918045044,
1980
+ "logps": -115.50324249267578,
1981
+ "loss": 0.1171,
1982
+ "objective": 0.1260891556739807,
1983
+ "ranking_idealized": 0.5333333611488342,
1984
+ "ranking_idealized_expo": 0.4749999940395355,
1985
+ "ranking_simple": 0.8916666507720947,
1986
+ "regularize": 0.1260891556739807,
1987
+ "step": 560
1988
+ },
1989
+ {
1990
+ "dpo_loss": 0.11428937315940857,
1991
+ "dpo_wo_beta": -0.142462819814682,
1992
+ "epoch": 1.6036844591402928,
1993
+ "grad_norm": 8.888993907970832,
1994
+ "learning_rate": 2.632215873827142e-06,
1995
+ "logits": -1.9027162790298462,
1996
+ "logps": -117.14366912841797,
1997
+ "loss": 0.1142,
1998
+ "objective": 0.11428937315940857,
1999
+ "ranking_idealized": 0.5416666865348816,
2000
+ "ranking_idealized_expo": 0.4583333432674408,
2001
+ "ranking_simple": 0.8916666507720947,
2002
+ "regularize": 0.11428937315940857,
2003
+ "step": 565
2004
+ },
2005
+ {
2006
+ "dpo_loss": 0.14770367741584778,
2007
+ "dpo_wo_beta": -0.5861695408821106,
2008
+ "epoch": 1.6178554558337268,
2009
+ "grad_norm": 9.396256315509223,
2010
+ "learning_rate": 2.5909207852394363e-06,
2011
+ "logits": -2.088587760925293,
2012
+ "logps": -113.1082992553711,
2013
+ "loss": 0.1098,
2014
+ "objective": 0.14770367741584778,
2015
+ "ranking_idealized": 0.6000000238418579,
2016
+ "ranking_idealized_expo": 0.5333333611488342,
2017
+ "ranking_simple": 0.9083333611488342,
2018
+ "regularize": 0.14770367741584778,
2019
+ "step": 570
2020
+ },
2021
+ {
2022
+ "dpo_loss": 0.1413314789533615,
2023
+ "dpo_wo_beta": -0.3796103298664093,
2024
+ "epoch": 1.632026452527161,
2025
+ "grad_norm": 16.325704474226345,
2026
+ "learning_rate": 2.5496008398168844e-06,
2027
+ "logits": -1.9472541809082031,
2028
+ "logps": -117.12679290771484,
2029
+ "loss": 0.1404,
2030
+ "objective": 0.1413314789533615,
2031
+ "ranking_idealized": 0.6583333611488342,
2032
+ "ranking_idealized_expo": 0.6166666746139526,
2033
+ "ranking_simple": 0.9166666865348816,
2034
+ "regularize": 0.1413314789533615,
2035
+ "step": 575
2036
+ },
2037
+ {
2038
+ "dpo_loss": 0.1715042144060135,
2039
+ "dpo_wo_beta": -0.48858124017715454,
2040
+ "epoch": 1.6461974492205953,
2041
+ "grad_norm": 11.97898567349363,
2042
+ "learning_rate": 2.508267334019988e-06,
2043
+ "logits": -1.935112476348877,
2044
+ "logps": -107.23208618164062,
2045
+ "loss": 0.1274,
2046
+ "objective": 0.1715042144060135,
2047
+ "ranking_idealized": 0.5583333373069763,
2048
+ "ranking_idealized_expo": 0.5,
2049
+ "ranking_simple": 0.8999999761581421,
2050
+ "regularize": 0.1715042144060135,
2051
+ "step": 580
2052
+ },
2053
+ {
2054
+ "epoch": 1.6547000472366555,
2055
+ "eval_dpo_loss": 0.8264312148094177,
2056
+ "eval_dpo_wo_beta": -5.2847466468811035,
2057
+ "eval_logits": -2.4379913806915283,
2058
+ "eval_logps": -119.59071350097656,
2059
+ "eval_loss": 0.808626651763916,
2060
+ "eval_objective": 0.8264312148094177,
2061
+ "eval_ranking_idealized": 0.6045548915863037,
2062
+ "eval_ranking_idealized_expo": 0.5279502868652344,
2063
+ "eval_ranking_simple": 0.5869565010070801,
2064
+ "eval_regularize": 0.8264312148094177,
2065
+ "eval_runtime": 355.6593,
2066
+ "eval_samples_per_second": 16.28,
2067
+ "eval_steps_per_second": 1.358,
2068
+ "step": 583
2069
+ },
2070
+ {
2071
+ "dpo_loss": 0.1341490000486374,
2072
+ "dpo_wo_beta": -0.5180007815361023,
2073
+ "epoch": 1.6603684459140293,
2074
+ "grad_norm": 15.27561747473412,
2075
+ "learning_rate": 2.46693156801652e-06,
2076
+ "logits": -1.8945667743682861,
2077
+ "logps": -110.8204116821289,
2078
+ "loss": 0.1227,
2079
+ "objective": 0.1341490000486374,
2080
+ "ranking_idealized": 0.6000000238418579,
2081
+ "ranking_idealized_expo": 0.5166666507720947,
2082
+ "ranking_simple": 0.9083333611488342,
2083
+ "regularize": 0.1341490000486374,
2084
+ "step": 585
2085
+ },
2086
+ {
2087
+ "dpo_loss": 0.13649246096611023,
2088
+ "dpo_wo_beta": -0.5407892465591431,
2089
+ "epoch": 1.6745394426074633,
2090
+ "grad_norm": 8.066407586221112,
2091
+ "learning_rate": 2.4256048425921693e-06,
2092
+ "logits": -1.8402847051620483,
2093
+ "logps": -117.5975112915039,
2094
+ "loss": 0.1263,
2095
+ "objective": 0.13649246096611023,
2096
+ "ranking_idealized": 0.6000000238418579,
2097
+ "ranking_idealized_expo": 0.46666666865348816,
2098
+ "ranking_simple": 0.9083333611488342,
2099
+ "regularize": 0.13649246096611023,
2100
+ "step": 590
2101
+ },
2102
+ {
2103
+ "dpo_loss": 0.14796380698680878,
2104
+ "dpo_wo_beta": -0.5909832119941711,
2105
+ "epoch": 1.6887104393008974,
2106
+ "grad_norm": 17.4246056389601,
2107
+ "learning_rate": 2.384298456061023e-06,
2108
+ "logits": -1.8368481397628784,
2109
+ "logps": -117.76559448242188,
2110
+ "loss": 0.1544,
2111
+ "objective": 0.14796380698680878,
2112
+ "ranking_idealized": 0.675000011920929,
2113
+ "ranking_idealized_expo": 0.5916666388511658,
2114
+ "ranking_simple": 0.8916666507720947,
2115
+ "regularize": 0.14796380698680878,
2116
+ "step": 595
2117
+ },
2118
+ {
2119
+ "dpo_loss": 0.15043622255325317,
2120
+ "dpo_wo_beta": -0.6607655882835388,
2121
+ "epoch": 1.7028814359943316,
2122
+ "grad_norm": 18.447927668502214,
2123
+ "learning_rate": 2.3430237011767166e-06,
2124
+ "logits": -1.839497447013855,
2125
+ "logps": -112.93590545654297,
2126
+ "loss": 0.15,
2127
+ "objective": 0.15043622255325317,
2128
+ "ranking_idealized": 0.6333333253860474,
2129
+ "ranking_idealized_expo": 0.550000011920929,
2130
+ "ranking_simple": 0.875,
2131
+ "regularize": 0.15043622255325317,
2132
+ "step": 600
2133
+ },
2134
+ {
2135
+ "dpo_loss": 0.11465544998645782,
2136
+ "dpo_wo_beta": -0.34068503975868225,
2137
+ "epoch": 1.7170524326877659,
2138
+ "grad_norm": 12.561303443006217,
2139
+ "learning_rate": 2.30179186204511e-06,
2140
+ "logits": -1.7405670881271362,
2141
+ "logps": -112.9049072265625,
2142
+ "loss": 0.1278,
2143
+ "objective": 0.11465544998645782,
2144
+ "ranking_idealized": 0.5583333373069763,
2145
+ "ranking_idealized_expo": 0.5,
2146
+ "ranking_simple": 0.8999999761581421,
2147
+ "regularize": 0.11465544998645782,
2148
+ "step": 605
2149
+ },
2150
+ {
2151
+ "dpo_loss": 0.1477740854024887,
2152
+ "dpo_wo_beta": -0.4079127609729767,
2153
+ "epoch": 1.7312234293811999,
2154
+ "grad_norm": 17.631548474702864,
2155
+ "learning_rate": 2.2606142110393248e-06,
2156
+ "logits": -1.7865701913833618,
2157
+ "logps": -109.32853698730469,
2158
+ "loss": 0.1331,
2159
+ "objective": 0.1477740854024887,
2160
+ "ranking_idealized": 0.5833333134651184,
2161
+ "ranking_idealized_expo": 0.5333333611488342,
2162
+ "ranking_simple": 0.875,
2163
+ "regularize": 0.1477740854024887,
2164
+ "step": 610
2165
+ },
2166
+ {
2167
+ "dpo_loss": 0.18279190361499786,
2168
+ "dpo_wo_beta": -0.8598226308822632,
2169
+ "epoch": 1.745394426074634,
2170
+ "grad_norm": 10.88386046876842,
2171
+ "learning_rate": 2.2195020057179897e-06,
2172
+ "logits": -1.7821184396743774,
2173
+ "logps": -111.59290313720703,
2174
+ "loss": 0.136,
2175
+ "objective": 0.18279190361499786,
2176
+ "ranking_idealized": 0.5833333134651184,
2177
+ "ranking_idealized_expo": 0.5166666507720947,
2178
+ "ranking_simple": 0.875,
2179
+ "regularize": 0.18279190361499786,
2180
+ "step": 615
2181
+ },
2182
+ {
2183
+ "dpo_loss": 0.1030309647321701,
2184
+ "dpo_wo_beta": -0.05606275424361229,
2185
+ "epoch": 1.759565422768068,
2186
+ "grad_norm": 12.949368950505932,
2187
+ "learning_rate": 2.1784664857475356e-06,
2188
+ "logits": -1.780458688735962,
2189
+ "logps": -108.14535522460938,
2190
+ "loss": 0.1303,
2191
+ "objective": 0.1030309647321701,
2192
+ "ranking_idealized": 0.6416666507720947,
2193
+ "ranking_idealized_expo": 0.5333333611488342,
2194
+ "ranking_simple": 0.9666666388511658,
2195
+ "regularize": 0.1030309647321701,
2196
+ "step": 620
2197
+ },
2198
+ {
2199
+ "dpo_loss": 0.158640518784523,
2200
+ "dpo_wo_beta": -0.4262932240962982,
2201
+ "epoch": 1.7737364194615022,
2202
+ "grad_norm": 15.36682532028932,
2203
+ "learning_rate": 2.1375188698293855e-06,
2204
+ "logits": -1.657003402709961,
2205
+ "logps": -110.36738586425781,
2206
+ "loss": 0.161,
2207
+ "objective": 0.158640518784523,
2208
+ "ranking_idealized": 0.6166666746139526,
2209
+ "ranking_idealized_expo": 0.5249999761581421,
2210
+ "ranking_simple": 0.9083333611488342,
2211
+ "regularize": 0.158640518784523,
2212
+ "step": 625
2213
+ },
2214
+ {
2215
+ "dpo_loss": 0.21892648935317993,
2216
+ "dpo_wo_beta": -0.9036411046981812,
2217
+ "epoch": 1.7879074161549362,
2218
+ "grad_norm": 14.959909753098716,
2219
+ "learning_rate": 2.096670352632873e-06,
2220
+ "logits": -1.7339377403259277,
2221
+ "logps": -107.9156723022461,
2222
+ "loss": 0.1673,
2223
+ "objective": 0.21892648935317993,
2224
+ "ranking_idealized": 0.5833333134651184,
2225
+ "ranking_idealized_expo": 0.49166667461395264,
2226
+ "ranking_simple": 0.7833333611488342,
2227
+ "regularize": 0.21892648935317993,
2228
+ "step": 630
2229
+ },
2230
+ {
2231
+ "dpo_loss": 0.14205454289913177,
2232
+ "dpo_wo_beta": -0.47540512681007385,
2233
+ "epoch": 1.8020784128483704,
2234
+ "grad_norm": 22.25702396524442,
2235
+ "learning_rate": 2.0559321017347286e-06,
2236
+ "logits": -2.000821352005005,
2237
+ "logps": -119.52375793457031,
2238
+ "loss": 0.1614,
2239
+ "objective": 0.14205454289913177,
2240
+ "ranking_idealized": 0.675000011920929,
2241
+ "ranking_idealized_expo": 0.5166666507720947,
2242
+ "ranking_simple": 0.8999999761581421,
2243
+ "regularize": 0.14205454289913177,
2244
+ "step": 635
2245
+ },
2246
+ {
2247
+ "epoch": 1.8049126121870571,
2248
+ "eval_dpo_loss": 0.8243346214294434,
2249
+ "eval_dpo_wo_beta": -5.281310081481934,
2250
+ "eval_logits": -2.4850430488586426,
2251
+ "eval_logps": -117.85846710205078,
2252
+ "eval_loss": 0.8209081292152405,
2253
+ "eval_objective": 0.8243346214294434,
2254
+ "eval_ranking_idealized": 0.6045548915863037,
2255
+ "eval_ranking_idealized_expo": 0.5279502868652344,
2256
+ "eval_ranking_simple": 0.5817805528640747,
2257
+ "eval_regularize": 0.8243346214294434,
2258
+ "eval_runtime": 351.818,
2259
+ "eval_samples_per_second": 16.457,
2260
+ "eval_steps_per_second": 1.373,
2261
+ "step": 636
2262
+ },
2263
+ {
2264
+ "dpo_loss": 0.13119691610336304,
2265
+ "dpo_wo_beta": -0.36706313490867615,
2266
+ "epoch": 1.8162494095418045,
2267
+ "grad_norm": 20.631621209893076,
2268
+ "learning_rate": 2.01531525456598e-06,
2269
+ "logits": -1.8479942083358765,
2270
+ "logps": -111.6270523071289,
2271
+ "loss": 0.1417,
2272
+ "objective": 0.13119691610336304,
2273
+ "ranking_idealized": 0.5249999761581421,
2274
+ "ranking_idealized_expo": 0.4833333194255829,
2275
+ "ranking_simple": 0.8583333492279053,
2276
+ "regularize": 0.13119691610336304,
2277
+ "step": 640
2278
+ },
2279
+ {
2280
+ "dpo_loss": 0.1571781039237976,
2281
+ "dpo_wo_beta": -0.5080724954605103,
2282
+ "epoch": 1.8304204062352385,
2283
+ "grad_norm": 20.508347622723388,
2284
+ "learning_rate": 1.974830915367086e-06,
2285
+ "logits": -1.9629262685775757,
2286
+ "logps": -122.33467102050781,
2287
+ "loss": 0.1732,
2288
+ "objective": 0.1571781039237976,
2289
+ "ranking_idealized": 0.6333333253860474,
2290
+ "ranking_idealized_expo": 0.5333333611488342,
2291
+ "ranking_simple": 0.875,
2292
+ "regularize": 0.1571781039237976,
2293
+ "step": 645
2294
+ },
2295
+ {
2296
+ "dpo_loss": 0.18711452186107635,
2297
+ "dpo_wo_beta": -0.5334885120391846,
2298
+ "epoch": 1.8445914029286725,
2299
+ "grad_norm": 17.78377983289135,
2300
+ "learning_rate": 1.93449015215215e-06,
2301
+ "logits": -1.960700511932373,
2302
+ "logps": -116.45585632324219,
2303
+ "loss": 0.1873,
2304
+ "objective": 0.18711452186107635,
2305
+ "ranking_idealized": 0.6333333253860474,
2306
+ "ranking_idealized_expo": 0.5,
2307
+ "ranking_simple": 0.875,
2308
+ "regularize": 0.18711452186107635,
2309
+ "step": 650
2310
+ },
2311
+ {
2312
+ "dpo_loss": 0.14089132845401764,
2313
+ "dpo_wo_beta": -0.257717490196228,
2314
+ "epoch": 1.8587623996221068,
2315
+ "grad_norm": 17.423684978791957,
2316
+ "learning_rate": 1.8943039936830347e-06,
2317
+ "logits": -1.7539128065109253,
2318
+ "logps": -105.96385192871094,
2319
+ "loss": 0.1703,
2320
+ "objective": 0.14089132845401764,
2321
+ "ranking_idealized": 0.5666666626930237,
2322
+ "ranking_idealized_expo": 0.46666666865348816,
2323
+ "ranking_simple": 0.8416666388511658,
2324
+ "regularize": 0.14089132845401764,
2325
+ "step": 655
2326
+ },
2327
+ {
2328
+ "dpo_loss": 0.12644439935684204,
2329
+ "dpo_wo_beta": -0.18994450569152832,
2330
+ "epoch": 1.872933396315541,
2331
+ "grad_norm": 14.114136095930363,
2332
+ "learning_rate": 1.8542834264542091e-06,
2333
+ "logits": -1.8740805387496948,
2334
+ "logps": -112.58109283447266,
2335
+ "loss": 0.1679,
2336
+ "objective": 0.12644439935684204,
2337
+ "ranking_idealized": 0.5666666626930237,
2338
+ "ranking_idealized_expo": 0.4833333194255829,
2339
+ "ranking_simple": 0.8833333253860474,
2340
+ "regularize": 0.12644439935684204,
2341
+ "step": 660
2342
+ },
2343
+ {
2344
+ "dpo_loss": 0.1596754938364029,
2345
+ "dpo_wo_beta": -0.5118387341499329,
2346
+ "epoch": 1.887104393008975,
2347
+ "grad_norm": 17.02657500840854,
2348
+ "learning_rate": 1.814439391689151e-06,
2349
+ "logits": -2.012057065963745,
2350
+ "logps": -106.6546401977539,
2351
+ "loss": 0.1748,
2352
+ "objective": 0.1596754938364029,
2353
+ "ranking_idealized": 0.6166666746139526,
2354
+ "ranking_idealized_expo": 0.5249999761581421,
2355
+ "ranking_simple": 0.8583333492279053,
2356
+ "regularize": 0.1596754938364029,
2357
+ "step": 665
2358
+ },
2359
+ {
2360
+ "dpo_loss": 0.1905670315027237,
2361
+ "dpo_wo_beta": -0.7486369609832764,
2362
+ "epoch": 1.901275389702409,
2363
+ "grad_norm": 14.652072743525036,
2364
+ "learning_rate": 1.7747827823491253e-06,
2365
+ "logits": -1.7807596921920776,
2366
+ "logps": -107.37975311279297,
2367
+ "loss": 0.1695,
2368
+ "objective": 0.1905670315027237,
2369
+ "ranking_idealized": 0.6000000238418579,
2370
+ "ranking_idealized_expo": 0.4833333194255829,
2371
+ "ranking_simple": 0.8666666746139526,
2372
+ "regularize": 0.1905670315027237,
2373
+ "step": 670
2374
+ },
2375
+ {
2376
+ "dpo_loss": 0.14669080078601837,
2377
+ "dpo_wo_beta": -0.3554477095603943,
2378
+ "epoch": 1.915446386395843,
2379
+ "grad_norm": 15.076428386388748,
2380
+ "learning_rate": 1.7353244401551566e-06,
2381
+ "logits": -1.8374218940734863,
2382
+ "logps": -107.03987884521484,
2383
+ "loss": 0.1555,
2384
+ "objective": 0.14669080078601837,
2385
+ "ranking_idealized": 0.6416666507720947,
2386
+ "ranking_idealized_expo": 0.5416666865348816,
2387
+ "ranking_simple": 0.8999999761581421,
2388
+ "regularize": 0.14669080078601837,
2389
+ "step": 675
2390
+ },
2391
+ {
2392
+ "dpo_loss": 0.15277433395385742,
2393
+ "dpo_wo_beta": -0.5579003095626831,
2394
+ "epoch": 1.9296173830892773,
2395
+ "grad_norm": 14.063016020977683,
2396
+ "learning_rate": 1.6960751526240122e-06,
2397
+ "logits": -1.9377697706222534,
2398
+ "logps": -118.40213775634766,
2399
+ "loss": 0.1487,
2400
+ "objective": 0.15277433395385742,
2401
+ "ranking_idealized": 0.625,
2402
+ "ranking_idealized_expo": 0.5416666865348816,
2403
+ "ranking_simple": 0.8833333253860474,
2404
+ "regularize": 0.15277433395385742,
2405
+ "step": 680
2406
+ },
2407
+ {
2408
+ "dpo_loss": 0.1468452513217926,
2409
+ "dpo_wo_beta": -0.34248843789100647,
2410
+ "epoch": 1.9437883797827114,
2411
+ "grad_norm": 20.037932580562952,
2412
+ "learning_rate": 1.6570456501189996e-06,
2413
+ "logits": -1.822561502456665,
2414
+ "logps": -116.04502868652344,
2415
+ "loss": 0.1616,
2416
+ "objective": 0.1468452513217926,
2417
+ "ranking_idealized": 0.5083333253860474,
2418
+ "ranking_idealized_expo": 0.46666666865348816,
2419
+ "ranking_simple": 0.8999999761581421,
2420
+ "regularize": 0.1468452513217926,
2421
+ "step": 685
2422
+ },
2423
+ {
2424
+ "epoch": 1.9551251771374587,
2425
+ "eval_dpo_loss": 0.857575535774231,
2426
+ "eval_dpo_wo_beta": -5.723405361175537,
2427
+ "eval_logits": -2.465576410293579,
2428
+ "eval_logps": -119.32209777832031,
2429
+ "eval_loss": 0.8382942080497742,
2430
+ "eval_objective": 0.857575535774231,
2431
+ "eval_ranking_idealized": 0.6045548915863037,
2432
+ "eval_ranking_idealized_expo": 0.5279502868652344,
2433
+ "eval_ranking_simple": 0.5797101259231567,
2434
+ "eval_regularize": 0.857575535774231,
2435
+ "eval_runtime": 359.3448,
2436
+ "eval_samples_per_second": 16.113,
2437
+ "eval_steps_per_second": 1.344,
2438
+ "step": 689
2439
+ },
2440
+ {
2441
+ "dpo_loss": 0.11434569954872131,
2442
+ "dpo_wo_beta": -0.045675624161958694,
2443
+ "epoch": 1.9579593764761456,
2444
+ "grad_norm": 13.437704226981893,
2445
+ "learning_rate": 1.6182466029163974e-06,
2446
+ "logits": -1.8100759983062744,
2447
+ "logps": -115.22444915771484,
2448
+ "loss": 0.1309,
2449
+ "objective": 0.11434569954872131,
2450
+ "ranking_idealized": 0.6166666746139526,
2451
+ "ranking_idealized_expo": 0.574999988079071,
2452
+ "ranking_simple": 0.8999999761581421,
2453
+ "regularize": 0.11434569954872131,
2454
+ "step": 690
2455
+ },
2456
+ {
2457
+ "dpo_loss": 0.19163179397583008,
2458
+ "dpo_wo_beta": -0.46952199935913086,
2459
+ "epoch": 1.9721303731695796,
2460
+ "grad_norm": 20.21550526964092,
2461
+ "learning_rate": 1.5796886182883053e-06,
2462
+ "logits": -1.829925775527954,
2463
+ "logps": -114.886962890625,
2464
+ "loss": 0.1753,
2465
+ "objective": 0.19163179397583008,
2466
+ "ranking_idealized": 0.5666666626930237,
2467
+ "ranking_idealized_expo": 0.4416666626930237,
2468
+ "ranking_simple": 0.8166666626930237,
2469
+ "regularize": 0.19163179397583008,
2470
+ "step": 695
2471
+ },
2472
+ {
2473
+ "dpo_loss": 0.18629121780395508,
2474
+ "dpo_wo_beta": -0.7477880120277405,
2475
+ "epoch": 1.9863013698630136,
2476
+ "grad_norm": 14.017066914132515,
2477
+ "learning_rate": 1.541382237602721e-06,
2478
+ "logits": -1.7919304370880127,
2479
+ "logps": -115.88526916503906,
2480
+ "loss": 0.176,
2481
+ "objective": 0.18629121780395508,
2482
+ "ranking_idealized": 0.5916666388511658,
2483
+ "ranking_idealized_expo": 0.5166666507720947,
2484
+ "ranking_simple": 0.8833333253860474,
2485
+ "regularize": 0.18629121780395508,
2486
+ "step": 700
2487
+ },
2488
+ {
2489
+ "dpo_loss": 0.20123900473117828,
2490
+ "dpo_wo_beta": -0.6124710440635681,
2491
+ "epoch": 2.0004723665564477,
2492
+ "grad_norm": 21.887245138739928,
2493
+ "learning_rate": 1.5033379334416376e-06,
2494
+ "logits": -1.786551594734192,
2495
+ "logps": -114.37857055664062,
2496
+ "loss": 0.2015,
2497
+ "objective": 0.20123900473117828,
2498
+ "ranking_idealized": 0.625,
2499
+ "ranking_idealized_expo": 0.5249999761581421,
2500
+ "ranking_simple": 0.8333333134651184,
2501
+ "regularize": 0.20123900473117828,
2502
+ "step": 705
2503
+ },
2504
+ {
2505
+ "dpo_loss": 0.08176574856042862,
2506
+ "dpo_wo_beta": -0.0886942520737648,
2507
+ "epoch": 2.0146433632498817,
2508
+ "grad_norm": 12.07978325252889,
2509
+ "learning_rate": 1.465566106737942e-06,
2510
+ "logits": -1.8880345821380615,
2511
+ "logps": -114.7125244140625,
2512
+ "loss": 0.1005,
2513
+ "objective": 0.08176574856042862,
2514
+ "ranking_idealized": 0.6000000238418579,
2515
+ "ranking_idealized_expo": 0.5249999761581421,
2516
+ "ranking_simple": 0.925000011920929,
2517
+ "regularize": 0.08176574856042862,
2518
+ "step": 710
2519
+ },
2520
+ {
2521
+ "dpo_loss": 0.07800193130970001,
2522
+ "dpo_wo_beta": -0.004813884384930134,
2523
+ "epoch": 2.028814359943316,
2524
+ "grad_norm": 10.892350166600437,
2525
+ "learning_rate": 1.4280770839319073e-06,
2526
+ "logits": -1.8223975896835327,
2527
+ "logps": -109.677001953125,
2528
+ "loss": 0.1064,
2529
+ "objective": 0.07800193130970001,
2530
+ "ranking_idealized": 0.6916666626930237,
2531
+ "ranking_idealized_expo": 0.5333333611488342,
2532
+ "ranking_simple": 0.949999988079071,
2533
+ "regularize": 0.07800193130970001,
2534
+ "step": 715
2535
+ },
2536
+ {
2537
+ "dpo_loss": 0.07249681651592255,
2538
+ "dpo_wo_beta": -0.03800208121538162,
2539
+ "epoch": 2.04298535663675,
2540
+ "grad_norm": 10.585775059236884,
2541
+ "learning_rate": 1.3908811141480408e-06,
2542
+ "logits": -1.7804607152938843,
2543
+ "logps": -119.00810241699219,
2544
+ "loss": 0.1011,
2545
+ "objective": 0.07249681651592255,
2546
+ "ranking_idealized": 0.6333333253860474,
2547
+ "ranking_idealized_expo": 0.5416666865348816,
2548
+ "ranking_simple": 0.949999988079071,
2549
+ "regularize": 0.07249681651592255,
2550
+ "step": 720
2551
+ },
2552
+ {
2553
+ "dpo_loss": 0.10758433490991592,
2554
+ "dpo_wo_beta": -0.4225389361381531,
2555
+ "epoch": 2.057156353330184,
2556
+ "grad_norm": 16.634608237356296,
2557
+ "learning_rate": 1.353988366393083e-06,
2558
+ "logits": -1.8837405443191528,
2559
+ "logps": -122.06383514404297,
2560
+ "loss": 0.1068,
2561
+ "objective": 0.10758433490991592,
2562
+ "ranking_idealized": 0.6499999761581421,
2563
+ "ranking_idealized_expo": 0.5583333373069763,
2564
+ "ranking_simple": 0.925000011920929,
2565
+ "regularize": 0.10758433490991592,
2566
+ "step": 725
2567
+ },
2568
+ {
2569
+ "dpo_loss": 0.16440601646900177,
2570
+ "dpo_wo_beta": -0.5969924926757812,
2571
+ "epoch": 2.0713273500236182,
2572
+ "grad_norm": 12.618281047842604,
2573
+ "learning_rate": 1.3174089267758983e-06,
2574
+ "logits": -1.7312003374099731,
2575
+ "logps": -118.56900024414062,
2576
+ "loss": 0.107,
2577
+ "objective": 0.16440601646900177,
2578
+ "ranking_idealized": 0.5166666507720947,
2579
+ "ranking_idealized_expo": 0.4333333373069763,
2580
+ "ranking_simple": 0.8583333492279053,
2581
+ "regularize": 0.16440601646900177,
2582
+ "step": 730
2583
+ },
2584
+ {
2585
+ "dpo_loss": 0.15623989701271057,
2586
+ "dpo_wo_beta": -0.9556598663330078,
2587
+ "epoch": 2.0854983467170523,
2588
+ "grad_norm": 16.05614648021729,
2589
+ "learning_rate": 1.2811527957500344e-06,
2590
+ "logits": -1.6055046319961548,
2591
+ "logps": -127.75144958496094,
2592
+ "loss": 0.1095,
2593
+ "objective": 0.15623989701271057,
2594
+ "ranking_idealized": 0.5583333373069763,
2595
+ "ranking_idealized_expo": 0.46666666865348816,
2596
+ "ranking_simple": 0.949999988079071,
2597
+ "regularize": 0.15623989701271057,
2598
+ "step": 735
2599
+ },
2600
+ {
2601
+ "dpo_loss": 0.11614324897527695,
2602
+ "dpo_wo_beta": -0.22688980400562286,
2603
+ "epoch": 2.0996693434104867,
2604
+ "grad_norm": 14.923960342956232,
2605
+ "learning_rate": 1.245229885379699e-06,
2606
+ "logits": -1.688416838645935,
2607
+ "logps": -123.08829498291016,
2608
+ "loss": 0.1063,
2609
+ "objective": 0.11614324897527695,
2610
+ "ranking_idealized": 0.6333333253860474,
2611
+ "ranking_idealized_expo": 0.5416666865348816,
2612
+ "ranking_simple": 0.949999988079071,
2613
+ "regularize": 0.11614324897527695,
2614
+ "step": 740
2615
+ },
2616
+ {
2617
+ "epoch": 2.10533774208786,
2618
+ "eval_dpo_loss": 0.9823706746101379,
2619
+ "eval_dpo_wo_beta": -7.331023216247559,
2620
+ "eval_logits": -2.2711641788482666,
2621
+ "eval_logps": -133.36373901367188,
2622
+ "eval_loss": 0.9485942721366882,
2623
+ "eval_objective": 0.9823706746101379,
2624
+ "eval_ranking_idealized": 0.6045548915863037,
2625
+ "eval_ranking_idealized_expo": 0.5279502868652344,
2626
+ "eval_ranking_simple": 0.5517598390579224,
2627
+ "eval_regularize": 0.9823706746101379,
2628
+ "eval_runtime": 364.9494,
2629
+ "eval_samples_per_second": 15.865,
2630
+ "eval_steps_per_second": 1.323,
2631
+ "step": 742
2632
+ },
2633
+ {
2634
+ "dpo_loss": 0.11820446699857712,
2635
+ "dpo_wo_beta": -0.5452965497970581,
2636
+ "epoch": 2.1138403401039207,
2637
+ "grad_norm": 14.405167306826879,
2638
+ "learning_rate": 1.2096500166298992e-06,
2639
+ "logits": -1.4971224069595337,
2640
+ "logps": -132.55892944335938,
2641
+ "loss": 0.1031,
2642
+ "objective": 0.11820446699857712,
2643
+ "ranking_idealized": 0.6916666626930237,
2644
+ "ranking_idealized_expo": 0.6166666746139526,
2645
+ "ranking_simple": 0.925000011920929,
2646
+ "regularize": 0.11820446699857712,
2647
+ "step": 745
2648
+ },
2649
+ {
2650
+ "dpo_loss": 0.08092837035655975,
2651
+ "dpo_wo_beta": -0.21483030915260315,
2652
+ "epoch": 2.1280113367973548,
2653
+ "grad_norm": 12.035987152428243,
2654
+ "learning_rate": 1.1744229166814889e-06,
2655
+ "logits": -1.698511004447937,
2656
+ "logps": -129.5937042236328,
2657
+ "loss": 0.0957,
2658
+ "objective": 0.08092837035655975,
2659
+ "ranking_idealized": 0.6666666865348816,
2660
+ "ranking_idealized_expo": 0.574999988079071,
2661
+ "ranking_simple": 0.9583333134651184,
2662
+ "regularize": 0.08092837035655975,
2663
+ "step": 750
2664
+ },
2665
+ {
2666
+ "dpo_loss": 0.09838299453258514,
2667
+ "dpo_wo_beta": -0.330010324716568,
2668
+ "epoch": 2.142182333490789,
2669
+ "grad_norm": 15.638085348810199,
2670
+ "learning_rate": 1.1395582162718524e-06,
2671
+ "logits": -1.6223360300064087,
2672
+ "logps": -128.86538696289062,
2673
+ "loss": 0.1147,
2674
+ "objective": 0.09838299453258514,
2675
+ "ranking_idealized": 0.5916666388511658,
2676
+ "ranking_idealized_expo": 0.4833333194255829,
2677
+ "ranking_simple": 0.925000011920929,
2678
+ "regularize": 0.09838299453258514,
2679
+ "step": 755
2680
+ },
2681
+ {
2682
+ "dpo_loss": 0.10984232276678085,
2683
+ "dpo_wo_beta": -0.2521561086177826,
2684
+ "epoch": 2.156353330184223,
2685
+ "grad_norm": 18.280317761955644,
2686
+ "learning_rate": 1.1050654470619602e-06,
2687
+ "logits": -1.6547772884368896,
2688
+ "logps": -118.33650970458984,
2689
+ "loss": 0.1127,
2690
+ "objective": 0.10984232276678085,
2691
+ "ranking_idealized": 0.5833333134651184,
2692
+ "ranking_idealized_expo": 0.5249999761581421,
2693
+ "ranking_simple": 0.9166666865348816,
2694
+ "regularize": 0.10984232276678085,
2695
+ "step": 760
2696
+ },
2697
+ {
2698
+ "dpo_loss": 0.11440528929233551,
2699
+ "dpo_wo_beta": -0.2931906580924988,
2700
+ "epoch": 2.170524326877657,
2701
+ "grad_norm": 12.536707104746414,
2702
+ "learning_rate": 1.0709540390305061e-06,
2703
+ "logits": -1.692717432975769,
2704
+ "logps": -118.69541931152344,
2705
+ "loss": 0.1215,
2706
+ "objective": 0.11440528929233551,
2707
+ "ranking_idealized": 0.5833333134651184,
2708
+ "ranking_idealized_expo": 0.5333333611488342,
2709
+ "ranking_simple": 0.9333333373069763,
2710
+ "regularize": 0.11440528929233551,
2711
+ "step": 765
2712
+ },
2713
+ {
2714
+ "dpo_loss": 0.06226298585534096,
2715
+ "dpo_wo_beta": -0.023008961230516434,
2716
+ "epoch": 2.1846953235710913,
2717
+ "grad_norm": 12.278836680753214,
2718
+ "learning_rate": 1.0372333178958462e-06,
2719
+ "logits": -1.8234201669692993,
2720
+ "logps": -122.00631713867188,
2721
+ "loss": 0.1046,
2722
+ "objective": 0.06226298585534096,
2723
+ "ranking_idealized": 0.5833333134651184,
2724
+ "ranking_idealized_expo": 0.49166667461395264,
2725
+ "ranking_simple": 0.9583333134651184,
2726
+ "regularize": 0.06226298585534096,
2727
+ "step": 770
2728
+ },
2729
+ {
2730
+ "dpo_loss": 0.13250760734081268,
2731
+ "dpo_wo_beta": -0.465701699256897,
2732
+ "epoch": 2.1988663202645253,
2733
+ "grad_norm": 18.99094104963921,
2734
+ "learning_rate": 1.0039125025664392e-06,
2735
+ "logits": -1.7803070545196533,
2736
+ "logps": -124.71762084960938,
2737
+ "loss": 0.1111,
2738
+ "objective": 0.13250760734081268,
2739
+ "ranking_idealized": 0.5416666865348816,
2740
+ "ranking_idealized_expo": 0.5083333253860474,
2741
+ "ranking_simple": 0.875,
2742
+ "regularize": 0.13250760734081268,
2743
+ "step": 775
2744
+ },
2745
+ {
2746
+ "dpo_loss": 0.13606421649456024,
2747
+ "dpo_wo_beta": -0.6162300705909729,
2748
+ "epoch": 2.2130373169579594,
2749
+ "grad_norm": 16.36095800479351,
2750
+ "learning_rate": 9.710007026204896e-07,
2751
+ "logits": -1.7376734018325806,
2752
+ "logps": -122.50430297851562,
2753
+ "loss": 0.1045,
2754
+ "objective": 0.13606421649456024,
2755
+ "ranking_idealized": 0.5,
2756
+ "ranking_idealized_expo": 0.4333333373069763,
2757
+ "ranking_simple": 0.8999999761581421,
2758
+ "regularize": 0.13606421649456024,
2759
+ "step": 780
2760
+ },
2761
+ {
2762
+ "dpo_loss": 0.11954029649496078,
2763
+ "dpo_wo_beta": -0.5017859935760498,
2764
+ "epoch": 2.2272083136513934,
2765
+ "grad_norm": 14.897356026126795,
2766
+ "learning_rate": 9.385069158154805e-07,
2767
+ "logits": -1.7108873128890991,
2768
+ "logps": -119.73731994628906,
2769
+ "loss": 0.1223,
2770
+ "objective": 0.11954029649496078,
2771
+ "ranking_idealized": 0.5583333373069763,
2772
+ "ranking_idealized_expo": 0.4583333432674408,
2773
+ "ranking_simple": 0.8916666507720947,
2774
+ "regularize": 0.11954029649496078,
2775
+ "step": 785
2776
+ },
2777
+ {
2778
+ "dpo_loss": 0.0877259224653244,
2779
+ "dpo_wo_beta": -0.3332770764827728,
2780
+ "epoch": 2.2413793103448274,
2781
+ "grad_norm": 14.08354106572375,
2782
+ "learning_rate": 9.064400256282757e-07,
2783
+ "logits": -1.7486475706100464,
2784
+ "logps": -122.89460754394531,
2785
+ "loss": 0.1063,
2786
+ "objective": 0.0877259224653244,
2787
+ "ranking_idealized": 0.625,
2788
+ "ranking_idealized_expo": 0.5666666626930237,
2789
+ "ranking_simple": 0.949999988079071,
2790
+ "regularize": 0.0877259224653244,
2791
+ "step": 790
2792
+ },
2793
+ {
2794
+ "dpo_loss": 0.09928978979587555,
2795
+ "dpo_wo_beta": -0.10460276901721954,
2796
+ "epoch": 2.255550307038262,
2797
+ "grad_norm": 15.557463813809257,
2798
+ "learning_rate": 8.74808798826467e-07,
2799
+ "logits": -1.8421998023986816,
2800
+ "logps": -120.40747833251953,
2801
+ "loss": 0.1017,
2802
+ "objective": 0.09928978979587555,
2803
+ "ranking_idealized": 0.699999988079071,
2804
+ "ranking_idealized_expo": 0.6000000238418579,
2805
+ "ranking_simple": 0.9416666626930237,
2806
+ "regularize": 0.09928978979587555,
2807
+ "step": 795
2808
+ },
2809
+ {
2810
+ "epoch": 2.255550307038262,
2811
+ "eval_dpo_loss": 0.8903655409812927,
2812
+ "eval_dpo_wo_beta": -6.205545902252197,
2813
+ "eval_logits": -2.4489927291870117,
2814
+ "eval_logps": -123.57449340820312,
2815
+ "eval_loss": 0.8711386919021606,
2816
+ "eval_objective": 0.8903655409812927,
2817
+ "eval_ranking_idealized": 0.6045548915863037,
2818
+ "eval_ranking_idealized_expo": 0.5279502868652344,
2819
+ "eval_ranking_simple": 0.5683229565620422,
2820
+ "eval_regularize": 0.8903655409812927,
2821
+ "eval_runtime": 347.0103,
2822
+ "eval_samples_per_second": 16.685,
2823
+ "eval_steps_per_second": 1.392,
2824
+ "step": 795
2825
+ },
2826
+ {
2827
+ "dpo_loss": 0.1340600550174713,
2828
+ "dpo_wo_beta": -0.5811701416969299,
2829
+ "epoch": 2.269721303731696,
2830
+ "grad_norm": 13.668309173603514,
2831
+ "learning_rate": 8.436218830716259e-07,
2832
+ "logits": -1.8454309701919556,
2833
+ "logps": -120.35511779785156,
2834
+ "loss": 0.1264,
2835
+ "objective": 0.1340600550174713,
2836
+ "ranking_idealized": 0.6499999761581421,
2837
+ "ranking_idealized_expo": 0.550000011920929,
2838
+ "ranking_simple": 0.925000011920929,
2839
+ "regularize": 0.1340600550174713,
2840
+ "step": 800
2841
+ },
2842
+ {
2843
+ "dpo_loss": 0.10816308110952377,
2844
+ "dpo_wo_beta": -0.1633211225271225,
2845
+ "epoch": 2.28389230042513,
2846
+ "grad_norm": 13.957508503091352,
2847
+ "learning_rate": 8.1288780455512e-07,
2848
+ "logits": -1.8811193704605103,
2849
+ "logps": -119.9056625366211,
2850
+ "loss": 0.1133,
2851
+ "objective": 0.10816308110952377,
2852
+ "ranking_idealized": 0.6000000238418579,
2853
+ "ranking_idealized_expo": 0.550000011920929,
2854
+ "ranking_simple": 0.9166666865348816,
2855
+ "regularize": 0.10816308110952377,
2856
+ "step": 805
2857
+ },
2858
+ {
2859
+ "dpo_loss": 0.1715787798166275,
2860
+ "dpo_wo_beta": -0.7547404766082764,
2861
+ "epoch": 2.298063297118564,
2862
+ "grad_norm": 12.765274097098457,
2863
+ "learning_rate": 7.826149656671386e-07,
2864
+ "logits": -1.9726245403289795,
2865
+ "logps": -114.58699798583984,
2866
+ "loss": 0.1146,
2867
+ "objective": 0.1715787798166275,
2868
+ "ranking_idealized": 0.6416666507720947,
2869
+ "ranking_idealized_expo": 0.5333333611488342,
2870
+ "ranking_simple": 0.9166666865348816,
2871
+ "regularize": 0.1715787798166275,
2872
+ "step": 810
2873
+ },
2874
+ {
2875
+ "dpo_loss": 0.09882104396820068,
2876
+ "dpo_wo_beta": -0.18077202141284943,
2877
+ "epoch": 2.312234293811998,
2878
+ "grad_norm": 11.818365173405272,
2879
+ "learning_rate": 7.528116426995605e-07,
2880
+ "logits": -1.7834192514419556,
2881
+ "logps": -118.5341796875,
2882
+ "loss": 0.0843,
2883
+ "objective": 0.09882104396820068,
2884
+ "ranking_idealized": 0.550000011920929,
2885
+ "ranking_idealized_expo": 0.4333333373069763,
2886
+ "ranking_simple": 0.8916666507720947,
2887
+ "regularize": 0.09882104396820068,
2888
+ "step": 815
2889
+ },
2890
+ {
2891
+ "dpo_loss": 0.1070082038640976,
2892
+ "dpo_wo_beta": -0.44529280066490173,
2893
+ "epoch": 2.3264052905054324,
2894
+ "grad_norm": 13.918285230217345,
2895
+ "learning_rate": 7.234859835833022e-07,
2896
+ "logits": -1.8069401979446411,
2897
+ "logps": -123.11463928222656,
2898
+ "loss": 0.1153,
2899
+ "objective": 0.1070082038640976,
2900
+ "ranking_idealized": 0.6083333492279053,
2901
+ "ranking_idealized_expo": 0.5583333373069763,
2902
+ "ranking_simple": 0.949999988079071,
2903
+ "regularize": 0.1070082038640976,
2904
+ "step": 820
2905
+ },
2906
+ {
2907
+ "dpo_loss": 0.14794430136680603,
2908
+ "dpo_wo_beta": -0.5907248258590698,
2909
+ "epoch": 2.3405762871988665,
2910
+ "grad_norm": 14.849580151366643,
2911
+ "learning_rate": 6.94646005660749e-07,
2912
+ "logits": -1.808493971824646,
2913
+ "logps": -116.64714050292969,
2914
+ "loss": 0.1107,
2915
+ "objective": 0.14794430136680603,
2916
+ "ranking_idealized": 0.6333333253860474,
2917
+ "ranking_idealized_expo": 0.5583333373069763,
2918
+ "ranking_simple": 0.9083333611488342,
2919
+ "regularize": 0.14794430136680603,
2920
+ "step": 825
2921
+ },
2922
+ {
2923
+ "dpo_loss": 0.08306514471769333,
2924
+ "dpo_wo_beta": -0.10000230371952057,
2925
+ "epoch": 2.3547472838923005,
2926
+ "grad_norm": 12.857101782602014,
2927
+ "learning_rate": 6.662995934939007e-07,
2928
+ "logits": -1.7857582569122314,
2929
+ "logps": -123.92733764648438,
2930
+ "loss": 0.1063,
2931
+ "objective": 0.08306514471769333,
2932
+ "ranking_idealized": 0.5833333134651184,
2933
+ "ranking_idealized_expo": 0.5166666507720947,
2934
+ "ranking_simple": 0.9583333134651184,
2935
+ "regularize": 0.08306514471769333,
2936
+ "step": 830
2937
+ },
2938
+ {
2939
+ "dpo_loss": 0.11827471107244492,
2940
+ "dpo_wo_beta": -0.40130358934402466,
2941
+ "epoch": 2.3689182805857345,
2942
+ "grad_norm": 13.825829745811577,
2943
+ "learning_rate": 6.384544967088063e-07,
2944
+ "logits": -1.8356945514678955,
2945
+ "logps": -122.48320770263672,
2946
+ "loss": 0.124,
2947
+ "objective": 0.11827471107244492,
2948
+ "ranking_idealized": 0.6000000238418579,
2949
+ "ranking_idealized_expo": 0.4583333432674408,
2950
+ "ranking_simple": 0.9333333373069763,
2951
+ "regularize": 0.11827471107244492,
2952
+ "step": 835
2953
+ },
2954
+ {
2955
+ "dpo_loss": 0.14169135689735413,
2956
+ "dpo_wo_beta": -0.3359481692314148,
2957
+ "epoch": 2.3830892772791685,
2958
+ "grad_norm": 18.243146391325514,
2959
+ "learning_rate": 6.111183278768956e-07,
2960
+ "logits": -1.8658840656280518,
2961
+ "logps": -123.42705535888672,
2962
+ "loss": 0.1321,
2963
+ "objective": 0.14169135689735413,
2964
+ "ranking_idealized": 0.5,
2965
+ "ranking_idealized_expo": 0.44999998807907104,
2966
+ "ranking_simple": 0.9083333611488342,
2967
+ "regularize": 0.14169135689735413,
2968
+ "step": 840
2969
+ },
2970
+ {
2971
+ "dpo_loss": 0.09381429105997086,
2972
+ "dpo_wo_beta": -0.14171645045280457,
2973
+ "epoch": 2.3972602739726026,
2974
+ "grad_norm": 24.500088347031785,
2975
+ "learning_rate": 5.842985604337769e-07,
2976
+ "logits": -1.7731019258499146,
2977
+ "logps": -125.81861877441406,
2978
+ "loss": 0.1225,
2979
+ "objective": 0.09381429105997086,
2980
+ "ranking_idealized": 0.6083333492279053,
2981
+ "ranking_idealized_expo": 0.5416666865348816,
2982
+ "ranking_simple": 0.8999999761581421,
2983
+ "regularize": 0.09381429105997086,
2984
+ "step": 845
2985
+ },
2986
+ {
2987
+ "epoch": 2.4057628719886632,
2988
+ "eval_dpo_loss": 0.9035148620605469,
2989
+ "eval_dpo_wo_beta": -6.352902889251709,
2990
+ "eval_logits": -2.4742591381073,
2991
+ "eval_logps": -124.53355407714844,
2992
+ "eval_loss": 0.882164478302002,
2993
+ "eval_objective": 0.9035148620605469,
2994
+ "eval_ranking_idealized": 0.6045548915863037,
2995
+ "eval_ranking_idealized_expo": 0.5279502868652344,
2996
+ "eval_ranking_simple": 0.5569358468055725,
2997
+ "eval_regularize": 0.9035148620605469,
2998
+ "eval_runtime": 366.478,
2999
+ "eval_samples_per_second": 15.799,
3000
+ "eval_steps_per_second": 1.318,
3001
+ "step": 848
3002
+ },
3003
+ {
3004
+ "dpo_loss": 0.10245585441589355,
3005
+ "dpo_wo_beta": -0.2030431628227234,
3006
+ "epoch": 2.413793103448276,
3007
+ "grad_norm": 22.363955547911008,
3008
+ "learning_rate": 5.580025266360764e-07,
3009
+ "logits": -1.7464776039123535,
3010
+ "logps": -122.80999755859375,
3011
+ "loss": 0.1449,
3012
+ "objective": 0.10245585441589355,
3013
+ "ranking_idealized": 0.6666666865348816,
3014
+ "ranking_idealized_expo": 0.5625,
3015
+ "ranking_simple": 0.9583333134651184,
3016
+ "regularize": 0.10245585441589355,
3017
+ "step": 850
3018
+ },
3019
+ {
3020
+ "dpo_loss": 0.14077231287956238,
3021
+ "dpo_wo_beta": -0.15207929909229279,
3022
+ "epoch": 2.42796410014171,
3023
+ "grad_norm": 14.848409925173506,
3024
+ "learning_rate": 5.322374155568688e-07,
3025
+ "logits": -1.8929237127304077,
3026
+ "logps": -115.12696838378906,
3027
+ "loss": 0.1124,
3028
+ "objective": 0.14077231287956238,
3029
+ "ranking_idealized": 0.5666666626930237,
3030
+ "ranking_idealized_expo": 0.5083333253860474,
3031
+ "ranking_simple": 0.9166666865348816,
3032
+ "regularize": 0.14077231287956238,
3033
+ "step": 855
3034
+ },
3035
+ {
3036
+ "dpo_loss": 0.1414971649646759,
3037
+ "dpo_wo_beta": -0.5974557995796204,
3038
+ "epoch": 2.442135096835144,
3039
+ "grad_norm": 11.710235822935351,
3040
+ "learning_rate": 5.070102711202606e-07,
3041
+ "logits": -1.7974507808685303,
3042
+ "logps": -121.47347259521484,
3043
+ "loss": 0.1083,
3044
+ "objective": 0.1414971649646759,
3045
+ "ranking_idealized": 0.6166666746139526,
3046
+ "ranking_idealized_expo": 0.5583333373069763,
3047
+ "ranking_simple": 0.9333333373069763,
3048
+ "regularize": 0.1414971649646759,
3049
+ "step": 860
3050
+ },
3051
+ {
3052
+ "dpo_loss": 0.09845638275146484,
3053
+ "dpo_wo_beta": -0.1998511403799057,
3054
+ "epoch": 2.4563060935285783,
3055
+ "grad_norm": 15.827446562598988,
3056
+ "learning_rate": 4.823279901756498e-07,
3057
+ "logits": -1.816353440284729,
3058
+ "logps": -122.7919692993164,
3059
+ "loss": 0.1063,
3060
+ "objective": 0.09845638275146484,
3061
+ "ranking_idealized": 0.5833333134651184,
3062
+ "ranking_idealized_expo": 0.5166666507720947,
3063
+ "ranking_simple": 0.9083333611488342,
3064
+ "regularize": 0.09845638275146484,
3065
+ "step": 865
3066
+ },
3067
+ {
3068
+ "dpo_loss": 0.11614971607923508,
3069
+ "dpo_wo_beta": -0.34934201836586,
3070
+ "epoch": 2.4704770902220123,
3071
+ "grad_norm": 15.618833316654502,
3072
+ "learning_rate": 4.581973206121948e-07,
3073
+ "logits": -1.9400283098220825,
3074
+ "logps": -119.73554992675781,
3075
+ "loss": 0.1153,
3076
+ "objective": 0.11614971607923508,
3077
+ "ranking_idealized": 0.5916666388511658,
3078
+ "ranking_idealized_expo": 0.5083333253860474,
3079
+ "ranking_simple": 0.925000011920929,
3080
+ "regularize": 0.11614971607923508,
3081
+ "step": 870
3082
+ },
3083
+ {
3084
+ "dpo_loss": 0.2065262645483017,
3085
+ "dpo_wo_beta": -1.1751881837844849,
3086
+ "epoch": 2.4846480869154464,
3087
+ "grad_norm": 17.223610041291963,
3088
+ "learning_rate": 4.3462485951401126e-07,
3089
+ "logits": -1.7437902688980103,
3090
+ "logps": -120.61251831054688,
3091
+ "loss": 0.1238,
3092
+ "objective": 0.2065262645483017,
3093
+ "ranking_idealized": 0.625,
3094
+ "ranking_idealized_expo": 0.574999988079071,
3095
+ "ranking_simple": 0.8666666746139526,
3096
+ "regularize": 0.2065262645483017,
3097
+ "step": 875
3098
+ },
3099
+ {
3100
+ "dpo_loss": 0.13995474576950073,
3101
+ "dpo_wo_beta": -0.27328214049339294,
3102
+ "epoch": 2.4988190836088804,
3103
+ "grad_norm": 11.937037662011573,
3104
+ "learning_rate": 4.116170513565942e-07,
3105
+ "logits": -1.9172199964523315,
3106
+ "logps": -115.04023742675781,
3107
+ "loss": 0.1093,
3108
+ "objective": 0.13995474576950073,
3109
+ "ranking_idealized": 0.5249999761581421,
3110
+ "ranking_idealized_expo": 0.49166667461395264,
3111
+ "ranking_simple": 0.8999999761581421,
3112
+ "regularize": 0.13995474576950073,
3113
+ "step": 880
3114
+ },
3115
+ {
3116
+ "dpo_loss": 0.13740381598472595,
3117
+ "dpo_wo_beta": -0.38727322220802307,
3118
+ "epoch": 2.5129900803023144,
3119
+ "grad_norm": 16.35266837824131,
3120
+ "learning_rate": 3.891801862449629e-07,
3121
+ "logits": -1.8533929586410522,
3122
+ "logps": -115.91497039794922,
3123
+ "loss": 0.1385,
3124
+ "objective": 0.13740381598472595,
3125
+ "ranking_idealized": 0.6083333492279053,
3126
+ "ranking_idealized_expo": 0.5416666865348816,
3127
+ "ranking_simple": 0.8916666507720947,
3128
+ "regularize": 0.13740381598472595,
3129
+ "step": 885
3130
+ },
3131
+ {
3132
+ "dpo_loss": 0.11444827914237976,
3133
+ "dpo_wo_beta": -0.31205496191978455,
3134
+ "epoch": 2.527161076995749,
3135
+ "grad_norm": 12.141683375579714,
3136
+ "learning_rate": 3.6732039819400686e-07,
3137
+ "logits": -1.6747931241989136,
3138
+ "logps": -116.25071716308594,
3139
+ "loss": 0.1173,
3140
+ "objective": 0.11444827914237976,
3141
+ "ranking_idealized": 0.5166666507720947,
3142
+ "ranking_idealized_expo": 0.4416666626930237,
3143
+ "ranking_simple": 0.9083333611488342,
3144
+ "regularize": 0.11444827914237976,
3145
+ "step": 890
3146
+ },
3147
+ {
3148
+ "dpo_loss": 0.21558310091495514,
3149
+ "dpo_wo_beta": -0.7720097303390503,
3150
+ "epoch": 2.541332073689183,
3151
+ "grad_norm": 17.5258932616616,
3152
+ "learning_rate": 3.46043663451511e-07,
3153
+ "logits": -1.938331127166748,
3154
+ "logps": -121.76246643066406,
3155
+ "loss": 0.1324,
3156
+ "objective": 0.21558310091495514,
3157
+ "ranking_idealized": 0.550000011920929,
3158
+ "ranking_idealized_expo": 0.4833333194255829,
3159
+ "ranking_simple": 0.8333333134651184,
3160
+ "regularize": 0.21558310091495514,
3161
+ "step": 895
3162
+ },
3163
+ {
3164
+ "dpo_loss": 0.1236240565776825,
3165
+ "dpo_wo_beta": -0.21698738634586334,
3166
+ "epoch": 2.555503070382617,
3167
+ "grad_norm": 16.69855766001795,
3168
+ "learning_rate": 3.253557988643072e-07,
3169
+ "logits": -1.8755207061767578,
3170
+ "logps": -119.14775848388672,
3171
+ "loss": 0.1157,
3172
+ "objective": 0.1236240565776825,
3173
+ "ranking_idealized": 0.625,
3174
+ "ranking_idealized_expo": 0.550000011920929,
3175
+ "ranking_simple": 0.9083333611488342,
3176
+ "regularize": 0.1236240565776825,
3177
+ "step": 900
3178
+ },
3179
+ {
3180
+ "epoch": 2.558337269721304,
3181
+ "eval_dpo_loss": 0.894111156463623,
3182
+ "eval_dpo_wo_beta": -6.213596343994141,
3183
+ "eval_logits": -2.4885809421539307,
3184
+ "eval_logps": -124.45829010009766,
3185
+ "eval_loss": 0.8717960715293884,
3186
+ "eval_objective": 0.894111156463623,
3187
+ "eval_ranking_idealized": 0.6045548915863037,
3188
+ "eval_ranking_idealized_expo": 0.5279502868652344,
3189
+ "eval_ranking_simple": 0.5621117949485779,
3190
+ "eval_regularize": 0.894111156463623,
3191
+ "eval_runtime": 342.6574,
3192
+ "eval_samples_per_second": 16.897,
3193
+ "eval_steps_per_second": 1.41,
3194
+ "step": 901
3195
+ },
3196
+ {
3197
+ "dpo_loss": 0.09204068034887314,
3198
+ "dpo_wo_beta": -0.17730669677257538,
3199
+ "epoch": 2.569674067076051,
3200
+ "grad_norm": 14.161016786369684,
3201
+ "learning_rate": 3.052624602880064e-07,
3202
+ "logits": -1.8424724340438843,
3203
+ "logps": -117.90782928466797,
3204
+ "loss": 0.0986,
3205
+ "objective": 0.09204068034887314,
3206
+ "ranking_idealized": 0.6000000238418579,
3207
+ "ranking_idealized_expo": 0.5333333611488342,
3208
+ "ranking_simple": 0.925000011920929,
3209
+ "regularize": 0.09204068034887314,
3210
+ "step": 905
3211
+ },
3212
+ {
3213
+ "dpo_loss": 0.11361932754516602,
3214
+ "dpo_wo_beta": -0.38244467973709106,
3215
+ "epoch": 2.583845063769485,
3216
+ "grad_norm": 12.421803606538058,
3217
+ "learning_rate": 2.8576914104074425e-07,
3218
+ "logits": -2.0089211463928223,
3219
+ "logps": -116.39904022216797,
3220
+ "loss": 0.1122,
3221
+ "objective": 0.11361932754516602,
3222
+ "ranking_idealized": 0.5916666388511658,
3223
+ "ranking_idealized_expo": 0.4583333432674408,
3224
+ "ranking_simple": 0.9166666865348816,
3225
+ "regularize": 0.11361932754516602,
3226
+ "step": 910
3227
+ },
3228
+ {
3229
+ "dpo_loss": 0.13471105694770813,
3230
+ "dpo_wo_beta": -0.5014829039573669,
3231
+ "epoch": 2.5980160604629194,
3232
+ "grad_norm": 15.12060209950672,
3233
+ "learning_rate": 2.6688117040136463e-07,
3234
+ "logits": -1.981037974357605,
3235
+ "logps": -121.86930084228516,
3236
+ "loss": 0.1153,
3237
+ "objective": 0.13471105694770813,
3238
+ "ranking_idealized": 0.625,
3239
+ "ranking_idealized_expo": 0.5166666507720947,
3240
+ "ranking_simple": 0.949999988079071,
3241
+ "regularize": 0.13471105694770813,
3242
+ "step": 915
3243
+ },
3244
+ {
3245
+ "dpo_loss": 0.08700807392597198,
3246
+ "dpo_wo_beta": -0.1876840889453888,
3247
+ "epoch": 2.6121870571563535,
3248
+ "grad_norm": 17.811975214160903,
3249
+ "learning_rate": 2.486037121524448e-07,
3250
+ "logits": -1.898934245109558,
3251
+ "logps": -120.13009643554688,
3252
+ "loss": 0.114,
3253
+ "objective": 0.08700807392597198,
3254
+ "ranking_idealized": 0.6333333253860474,
3255
+ "ranking_idealized_expo": 0.550000011920929,
3256
+ "ranking_simple": 0.9416666626930237,
3257
+ "regularize": 0.08700807392597198,
3258
+ "step": 920
3259
+ },
3260
+ {
3261
+ "dpo_loss": 0.0918925479054451,
3262
+ "dpo_wo_beta": -0.14420188963413239,
3263
+ "epoch": 2.6263580538497875,
3264
+ "grad_norm": 16.6284598820075,
3265
+ "learning_rate": 2.3094176316856982e-07,
3266
+ "logits": -1.8268101215362549,
3267
+ "logps": -121.2860107421875,
3268
+ "loss": 0.1064,
3269
+ "objective": 0.0918925479054451,
3270
+ "ranking_idealized": 0.550000011920929,
3271
+ "ranking_idealized_expo": 0.4749999940395355,
3272
+ "ranking_simple": 0.9083333611488342,
3273
+ "regularize": 0.0918925479054451,
3274
+ "step": 925
3275
+ },
3276
+ {
3277
+ "dpo_loss": 0.15886452794075012,
3278
+ "dpo_wo_beta": -0.7213179469108582,
3279
+ "epoch": 2.6405290505432215,
3280
+ "grad_norm": 15.32952011074324,
3281
+ "learning_rate": 2.13900152050239e-07,
3282
+ "logits": -1.9606980085372925,
3283
+ "logps": -110.05363464355469,
3284
+ "loss": 0.1179,
3285
+ "objective": 0.15886452794075012,
3286
+ "ranking_idealized": 0.5416666865348816,
3287
+ "ranking_idealized_expo": 0.4833333194255829,
3288
+ "ranking_simple": 0.8999999761581421,
3289
+ "regularize": 0.15886452794075012,
3290
+ "step": 930
3291
+ },
3292
+ {
3293
+ "dpo_loss": 0.13784056901931763,
3294
+ "dpo_wo_beta": -0.5006576776504517,
3295
+ "epoch": 2.6547000472366555,
3296
+ "grad_norm": 14.658353796887713,
3297
+ "learning_rate": 1.9748353780377234e-07,
3298
+ "logits": -1.9395031929016113,
3299
+ "logps": -119.9269790649414,
3300
+ "loss": 0.1254,
3301
+ "objective": 0.13784056901931763,
3302
+ "ranking_idealized": 0.6583333611488342,
3303
+ "ranking_idealized_expo": 0.6000000238418579,
3304
+ "ranking_simple": 0.8916666507720947,
3305
+ "regularize": 0.13784056901931763,
3306
+ "step": 935
3307
+ },
3308
+ {
3309
+ "dpo_loss": 0.12213913351297379,
3310
+ "dpo_wo_beta": -0.43640393018722534,
3311
+ "epoch": 2.66887104393009,
3312
+ "grad_norm": 14.844912954939305,
3313
+ "learning_rate": 1.8169640856758652e-07,
3314
+ "logits": -1.952646017074585,
3315
+ "logps": -121.75474548339844,
3316
+ "loss": 0.1117,
3317
+ "objective": 0.12213913351297379,
3318
+ "ranking_idealized": 0.6000000238418579,
3319
+ "ranking_idealized_expo": 0.5083333253860474,
3320
+ "ranking_simple": 0.9166666865348816,
3321
+ "regularize": 0.12213913351297379,
3322
+ "step": 940
3323
+ },
3324
+ {
3325
+ "dpo_loss": 0.1136295348405838,
3326
+ "dpo_wo_beta": -0.17752434313297272,
3327
+ "epoch": 2.6830420406235236,
3328
+ "grad_norm": 13.220834082246482,
3329
+ "learning_rate": 1.6654308038518057e-07,
3330
+ "logits": -1.7970060110092163,
3331
+ "logps": -117.90103149414062,
3332
+ "loss": 0.1054,
3333
+ "objective": 0.1136295348405838,
3334
+ "ranking_idealized": 0.5833333134651184,
3335
+ "ranking_idealized_expo": 0.49166667461395264,
3336
+ "ranking_simple": 0.9083333611488342,
3337
+ "regularize": 0.1136295348405838,
3338
+ "step": 945
3339
+ },
3340
+ {
3341
+ "dpo_loss": 0.16729401051998138,
3342
+ "dpo_wo_beta": -0.7473469972610474,
3343
+ "epoch": 2.697213037316958,
3344
+ "grad_norm": 14.369145851061829,
3345
+ "learning_rate": 1.5202769602517514e-07,
3346
+ "logits": -1.8816020488739014,
3347
+ "logps": -115.14582061767578,
3348
+ "loss": 0.1387,
3349
+ "objective": 0.16729401051998138,
3350
+ "ranking_idealized": 0.625,
3351
+ "ranking_idealized_expo": 0.5,
3352
+ "ranking_simple": 0.8999999761581421,
3353
+ "regularize": 0.16729401051998138,
3354
+ "step": 950
3355
+ },
3356
+ {
3357
+ "epoch": 2.708549834671705,
3358
+ "eval_dpo_loss": 0.8891981840133667,
3359
+ "eval_dpo_wo_beta": -6.16098690032959,
3360
+ "eval_logits": -2.508572816848755,
3361
+ "eval_logps": -123.41434478759766,
3362
+ "eval_loss": 0.8687644004821777,
3363
+ "eval_objective": 0.8891981840133667,
3364
+ "eval_ranking_idealized": 0.6045548915863037,
3365
+ "eval_ranking_idealized_expo": 0.5279502868652344,
3366
+ "eval_ranking_simple": 0.5579710006713867,
3367
+ "eval_regularize": 0.8891981840133667,
3368
+ "eval_runtime": 379.9179,
3369
+ "eval_samples_per_second": 15.24,
3370
+ "eval_steps_per_second": 1.271,
3371
+ "step": 954
3372
+ },
3373
+ {
3374
+ "dpo_loss": 0.16232462227344513,
3375
+ "dpo_wo_beta": -0.5988053679466248,
3376
+ "epoch": 2.711384034010392,
3377
+ "grad_norm": 18.96223814657741,
3378
+ "learning_rate": 1.381542238487188e-07,
3379
+ "logits": -1.8838648796081543,
3380
+ "logps": -119.06771087646484,
3381
+ "loss": 0.1298,
3382
+ "objective": 0.16232462227344513,
3383
+ "ranking_idealized": 0.5833333134651184,
3384
+ "ranking_idealized_expo": 0.5083333253860474,
3385
+ "ranking_simple": 0.8999999761581421,
3386
+ "regularize": 0.16232462227344513,
3387
+ "step": 955
3388
+ },
3389
+ {
3390
+ "dpo_loss": 0.14441066980361938,
3391
+ "dpo_wo_beta": -0.4333815276622772,
3392
+ "epoch": 2.725555030703826,
3393
+ "grad_norm": 13.716940458471806,
3394
+ "learning_rate": 1.2492645672457838e-07,
3395
+ "logits": -2.032045364379883,
3396
+ "logps": -113.05756378173828,
3397
+ "loss": 0.1189,
3398
+ "objective": 0.14441066980361938,
3399
+ "ranking_idealized": 0.6000000238418579,
3400
+ "ranking_idealized_expo": 0.5083333253860474,
3401
+ "ranking_simple": 0.8999999761581421,
3402
+ "regularize": 0.14441066980361938,
3403
+ "step": 960
3404
+ },
3405
+ {
3406
+ "dpo_loss": 0.0981813296675682,
3407
+ "dpo_wo_beta": -0.19802381098270416,
3408
+ "epoch": 2.73972602739726,
3409
+ "grad_norm": 15.207158850260772,
3410
+ "learning_rate": 1.1234801099220787e-07,
3411
+ "logits": -1.7988998889923096,
3412
+ "logps": -122.7618408203125,
3413
+ "loss": 0.1111,
3414
+ "objective": 0.0981813296675682,
3415
+ "ranking_idealized": 0.625,
3416
+ "ranking_idealized_expo": 0.5666666626930237,
3417
+ "ranking_simple": 0.9333333373069763,
3418
+ "regularize": 0.0981813296675682,
3419
+ "step": 965
3420
+ },
3421
+ {
3422
+ "dpo_loss": 0.1318301260471344,
3423
+ "dpo_wo_beta": -0.2933843731880188,
3424
+ "epoch": 2.753897024090694,
3425
+ "grad_norm": 19.758130540339153,
3426
+ "learning_rate": 1.004223254730749e-07,
3427
+ "logits": -1.7169368267059326,
3428
+ "logps": -120.43925476074219,
3429
+ "loss": 0.1278,
3430
+ "objective": 0.1318301260471344,
3431
+ "ranking_idealized": 0.6083333492279053,
3432
+ "ranking_idealized_expo": 0.4833333194255829,
3433
+ "ranking_simple": 0.8916666507720947,
3434
+ "regularize": 0.1318301260471344,
3435
+ "step": 970
3436
+ },
3437
+ {
3438
+ "dpo_loss": 0.058168552815914154,
3439
+ "dpo_wo_beta": -8.394511678488925e-05,
3440
+ "epoch": 2.7680680207841286,
3441
+ "grad_norm": 12.662905511630496,
3442
+ "learning_rate": 8.915266053052374e-08,
3443
+ "logits": -1.902711033821106,
3444
+ "logps": -116.11229705810547,
3445
+ "loss": 0.0999,
3446
+ "objective": 0.058168552815914154,
3447
+ "ranking_idealized": 0.6333333253860474,
3448
+ "ranking_idealized_expo": 0.5,
3449
+ "ranking_simple": 0.925000011920929,
3450
+ "regularize": 0.058168552815914154,
3451
+ "step": 975
3452
+ },
3453
+ {
3454
+ "dpo_loss": 0.10343047231435776,
3455
+ "dpo_wo_beta": -0.17137210071086884,
3456
+ "epoch": 2.7822390174775626,
3457
+ "grad_norm": 18.09899373584344,
3458
+ "learning_rate": 7.854209717842231e-08,
3459
+ "logits": -1.8915067911148071,
3460
+ "logps": -118.23885345458984,
3461
+ "loss": 0.1108,
3462
+ "objective": 0.10343047231435776,
3463
+ "ranking_idealized": 0.5416666865348816,
3464
+ "ranking_idealized_expo": 0.4416666626930237,
3465
+ "ranking_simple": 0.949999988079071,
3466
+ "regularize": 0.10343047231435776,
3467
+ "step": 980
3468
+ },
3469
+ {
3470
+ "dpo_loss": 0.11380515992641449,
3471
+ "dpo_wo_beta": -0.2935677468776703,
3472
+ "epoch": 2.7964100141709967,
3473
+ "grad_norm": 15.185634936609162,
3474
+ "learning_rate": 6.859353623884569e-08,
3475
+ "logits": -1.851272463798523,
3476
+ "logps": -114.0352783203125,
3477
+ "loss": 0.1005,
3478
+ "objective": 0.11380515992641449,
3479
+ "ranking_idealized": 0.625,
3480
+ "ranking_idealized_expo": 0.5166666507720947,
3481
+ "ranking_simple": 0.8833333253860474,
3482
+ "regularize": 0.11380515992641449,
3483
+ "step": 985
3484
+ },
3485
+ {
3486
+ "dpo_loss": 0.10726428776979446,
3487
+ "dpo_wo_beta": -0.24069656431674957,
3488
+ "epoch": 2.8105810108644307,
3489
+ "grad_norm": 16.660819641239655,
3490
+ "learning_rate": 5.930969754901844e-08,
3491
+ "logits": -1.8860033750534058,
3492
+ "logps": -116.69475555419922,
3493
+ "loss": 0.1285,
3494
+ "objective": 0.10726428776979446,
3495
+ "ranking_idealized": 0.49166667461395264,
3496
+ "ranking_idealized_expo": 0.4333333373069763,
3497
+ "ranking_simple": 0.8833333253860474,
3498
+ "regularize": 0.10726428776979446,
3499
+ "step": 990
3500
+ },
3501
+ {
3502
+ "dpo_loss": 0.12048947066068649,
3503
+ "dpo_wo_beta": -0.4086553454399109,
3504
+ "epoch": 2.8247520075578647,
3505
+ "grad_norm": 18.459518556479477,
3506
+ "learning_rate": 5.069311921774039e-08,
3507
+ "logits": -1.8980218172073364,
3508
+ "logps": -120.89018249511719,
3509
+ "loss": 0.1282,
3510
+ "objective": 0.12048947066068649,
3511
+ "ranking_idealized": 0.6416666507720947,
3512
+ "ranking_idealized_expo": 0.550000011920929,
3513
+ "ranking_simple": 0.925000011920929,
3514
+ "regularize": 0.12048947066068649,
3515
+ "step": 995
3516
+ },
3517
+ {
3518
+ "dpo_loss": 0.08448319137096405,
3519
+ "dpo_wo_beta": -0.17254652082920074,
3520
+ "epoch": 2.838923004251299,
3521
+ "grad_norm": 13.073827319036173,
3522
+ "learning_rate": 4.2746156931490756e-08,
3523
+ "logits": -1.8106515407562256,
3524
+ "logps": -113.39408874511719,
3525
+ "loss": 0.1237,
3526
+ "objective": 0.08448319137096405,
3527
+ "ranking_idealized": 0.6000000238418579,
3528
+ "ranking_idealized_expo": 0.49166667461395264,
3529
+ "ranking_simple": 0.9333333373069763,
3530
+ "regularize": 0.08448319137096405,
3531
+ "step": 1000
3532
+ },
3533
+ {
3534
+ "dpo_loss": 0.12625885009765625,
3535
+ "dpo_wo_beta": -0.28751522302627563,
3536
+ "epoch": 2.853094000944733,
3537
+ "grad_norm": 11.352338165734487,
3538
+ "learning_rate": 3.547098331040916e-08,
3539
+ "logits": -1.8715885877609253,
3540
+ "logps": -115.85346984863281,
3541
+ "loss": 0.1219,
3542
+ "objective": 0.12625885009765625,
3543
+ "ranking_idealized": 0.5833333134651184,
3544
+ "ranking_idealized_expo": 0.4749999940395355,
3545
+ "ranking_simple": 0.8833333253860474,
3546
+ "regularize": 0.12625885009765625,
3547
+ "step": 1005
3548
+ },
3549
+ {
3550
+ "epoch": 2.8587623996221065,
3551
+ "eval_dpo_loss": 0.8881875872612,
3552
+ "eval_dpo_wo_beta": -6.153732776641846,
3553
+ "eval_logits": -2.5127134323120117,
3554
+ "eval_logps": -123.14542388916016,
3555
+ "eval_loss": 0.868183434009552,
3556
+ "eval_objective": 0.8881875872612,
3557
+ "eval_ranking_idealized": 0.6045548915863037,
3558
+ "eval_ranking_idealized_expo": 0.5279502868652344,
3559
+ "eval_ranking_simple": 0.5600414276123047,
3560
+ "eval_regularize": 0.8881875872612,
3561
+ "eval_runtime": 375.0227,
3562
+ "eval_samples_per_second": 15.439,
3563
+ "eval_steps_per_second": 1.288,
3564
+ "step": 1007
3565
+ },
3566
+ {
3567
+ "dpo_loss": 0.12578138709068298,
3568
+ "dpo_wo_beta": -0.4055772125720978,
3569
+ "epoch": 2.8672649976381672,
3570
+ "grad_norm": 13.654143333766829,
3571
+ "learning_rate": 2.8869587314321324e-08,
3572
+ "logits": -1.871021032333374,
3573
+ "logps": -118.54988861083984,
3574
+ "loss": 0.1301,
3575
+ "objective": 0.12578138709068298,
3576
+ "ranking_idealized": 0.6499999761581421,
3577
+ "ranking_idealized_expo": 0.5666666626930237,
3578
+ "ranking_simple": 0.9166666865348816,
3579
+ "regularize": 0.12578138709068298,
3580
+ "step": 1010
3581
+ },
3582
+ {
3583
+ "dpo_loss": 0.11141829192638397,
3584
+ "dpo_wo_beta": -0.452913373708725,
3585
+ "epoch": 2.8814359943316012,
3586
+ "grad_norm": 14.00967768494451,
3587
+ "learning_rate": 2.2943773698977935e-08,
3588
+ "logits": -1.8538991212844849,
3589
+ "logps": -119.40221405029297,
3590
+ "loss": 0.1157,
3591
+ "objective": 0.11141829192638397,
3592
+ "ranking_idealized": 0.6000000238418579,
3593
+ "ranking_idealized_expo": 0.5,
3594
+ "ranking_simple": 0.9416666626930237,
3595
+ "regularize": 0.11141829192638397,
3596
+ "step": 1015
3597
+ },
3598
+ {
3599
+ "dpo_loss": 0.10787668824195862,
3600
+ "dpo_wo_beta": -0.35029396414756775,
3601
+ "epoch": 2.8956069910250353,
3602
+ "grad_norm": 12.590503217808422,
3603
+ "learning_rate": 1.7695162522652352e-08,
3604
+ "logits": -1.9000986814498901,
3605
+ "logps": -122.90519714355469,
3606
+ "loss": 0.1076,
3607
+ "objective": 0.10787668824195862,
3608
+ "ranking_idealized": 0.6083333492279053,
3609
+ "ranking_idealized_expo": 0.5,
3610
+ "ranking_simple": 0.8999999761581421,
3611
+ "regularize": 0.10787668824195862,
3612
+ "step": 1020
3613
+ },
3614
+ {
3615
+ "dpo_loss": 0.11394120752811432,
3616
+ "dpo_wo_beta": -0.36261746287345886,
3617
+ "epoch": 2.9097779877184697,
3618
+ "grad_norm": 10.606906345600125,
3619
+ "learning_rate": 1.3125188703233815e-08,
3620
+ "logits": -1.8986002206802368,
3621
+ "logps": -116.403564453125,
3622
+ "loss": 0.1089,
3623
+ "objective": 0.11394120752811432,
3624
+ "ranking_idealized": 0.5583333373069763,
3625
+ "ranking_idealized_expo": 0.44999998807907104,
3626
+ "ranking_simple": 0.9333333373069763,
3627
+ "regularize": 0.11394120752811432,
3628
+ "step": 1025
3629
+ },
3630
+ {
3631
+ "dpo_loss": 0.0906638652086258,
3632
+ "dpo_wo_beta": -0.1376449316740036,
3633
+ "epoch": 2.9239489844119038,
3634
+ "grad_norm": 12.750602018189479,
3635
+ "learning_rate": 9.235101625932885e-09,
3636
+ "logits": -2.033400058746338,
3637
+ "logps": -113.65220642089844,
3638
+ "loss": 0.1197,
3639
+ "objective": 0.0906638652086258,
3640
+ "ranking_idealized": 0.6000000238418579,
3641
+ "ranking_idealized_expo": 0.5249999761581421,
3642
+ "ranking_simple": 0.925000011920929,
3643
+ "regularize": 0.0906638652086258,
3644
+ "step": 1030
3645
+ },
3646
+ {
3647
+ "dpo_loss": 0.10265343636274338,
3648
+ "dpo_wo_beta": -0.1061137467622757,
3649
+ "epoch": 2.938119981105338,
3650
+ "grad_norm": 15.296605965797069,
3651
+ "learning_rate": 6.025964801714412e-09,
3652
+ "logits": -1.8468897342681885,
3653
+ "logps": -119.85134887695312,
3654
+ "loss": 0.1049,
3655
+ "objective": 0.10265343636274338,
3656
+ "ranking_idealized": 0.6083333492279053,
3657
+ "ranking_idealized_expo": 0.49166667461395264,
3658
+ "ranking_simple": 0.8916666507720947,
3659
+ "regularize": 0.10265343636274338,
3660
+ "step": 1035
3661
+ },
3662
+ {
3663
+ "dpo_loss": 0.09396873414516449,
3664
+ "dpo_wo_beta": -0.1912200003862381,
3665
+ "epoch": 2.952290977798772,
3666
+ "grad_norm": 15.90972962002085,
3667
+ "learning_rate": 3.4986555765434415e-09,
3668
+ "logits": -1.8800926208496094,
3669
+ "logps": -122.51961517333984,
3670
+ "loss": 0.0994,
3671
+ "objective": 0.09396873414516449,
3672
+ "ranking_idealized": 0.6499999761581421,
3673
+ "ranking_idealized_expo": 0.5583333373069763,
3674
+ "ranking_simple": 0.9083333611488342,
3675
+ "regularize": 0.09396873414516449,
3676
+ "step": 1040
3677
+ },
3678
+ {
3679
+ "dpo_loss": 0.14692950248718262,
3680
+ "dpo_wo_beta": -0.6586350798606873,
3681
+ "epoch": 2.966461974492206,
3682
+ "grad_norm": 8.0573279067109,
3683
+ "learning_rate": 1.6538648915270794e-09,
3684
+ "logits": -1.8756026029586792,
3685
+ "logps": -119.65303039550781,
3686
+ "loss": 0.1082,
3687
+ "objective": 0.14692950248718262,
3688
+ "ranking_idealized": 0.6416666507720947,
3689
+ "ranking_idealized_expo": 0.5416666865348816,
3690
+ "ranking_simple": 0.9166666865348816,
3691
+ "regularize": 0.14692950248718262,
3692
+ "step": 1045
3693
+ },
3694
+ {
3695
+ "dpo_loss": 0.07982174307107925,
3696
+ "dpo_wo_beta": -0.0584401935338974,
3697
+ "epoch": 2.9806329711856403,
3698
+ "grad_norm": 9.88886611903832,
3699
+ "learning_rate": 4.920970940180958e-10,
3700
+ "logits": -1.912126898765564,
3701
+ "logps": -116.61032104492188,
3702
+ "loss": 0.0891,
3703
+ "objective": 0.07982174307107925,
3704
+ "ranking_idealized": 0.6166666746139526,
3705
+ "ranking_idealized_expo": 0.5249999761581421,
3706
+ "ranking_simple": 0.925000011920929,
3707
+ "regularize": 0.07982174307107925,
3708
+ "step": 1050
3709
+ },
3710
+ {
3711
+ "dpo_loss": 0.07663024961948395,
3712
+ "dpo_wo_beta": -0.015355088748037815,
3713
+ "epoch": 2.9948039678790743,
3714
+ "grad_norm": 13.025122561532541,
3715
+ "learning_rate": 1.3669799732163314e-11,
3716
+ "logits": -1.775391697883606,
3717
+ "logps": -116.977294921875,
3718
+ "loss": 0.0869,
3719
+ "objective": 0.07663024961948395,
3720
+ "ranking_idealized": 0.6333333253860474,
3721
+ "ranking_idealized_expo": 0.550000011920929,
3722
+ "ranking_simple": 0.8999999761581421,
3723
+ "regularize": 0.07663024961948395,
3724
+ "step": 1055
3725
+ },
3726
  {
3727
  "epoch": 2.9976381672177608,
3728
+ "step": 1056,
3729
  "total_flos": 0.0,
3730
+ "train_loss": 0.022545777056648425,
3731
+ "train_runtime": 4386.5835,
3732
+ "train_samples_per_second": 34.744,
3733
+ "train_steps_per_second": 0.241
3734
  }
3735
  ],
3736
  "logging_steps": 5,
3737
+ "max_steps": 1056,
3738
  "num_input_tokens_seen": 0,
3739
  "num_train_epochs": 3,
3740
  "save_steps": 53,