Hugging Face's logo

hugodk-sch
/
aftonposten-6b-align-scan

PEFT
Safetensors
gptj
alignment-handbook
trl
dpo
Generated from Trainer
4-bit precision
bitsandbytes
Model card Files Community
hugodk-sch commited on
Commit
9c4c3a6
•
1 Parent(s): e3d586a

Training in progress, step 1300

Browse files
Files changed (8) hide show
  1. README.md +15 -27
  2. adapter_config.json +4 -4
  3. adapter_model.safetensors +1 -1
  4. all_results.json +15 -15
  5. eval_results.json +11 -11
  6. train_results.json +5 -5
  7. trainer_state.json +0 -0
  8. training_args.bin +1 -1
README.md CHANGED
@@ -20,15 +20,15 @@ should probably proofread and complete it, then remove this comment. -->
20
 
21
  This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
22
  It achieves the following results on the evaluation set:
23
- - Loss: 0.4837
24
- - Rewards/chosen: 0.1032
25
- - Rewards/rejected: 0.0249
26
- - Rewards/accuracies: 0.6034
27
- - Rewards/margins: 0.0783
28
- - Logps/rejected: -37.4336
29
- - Logps/chosen: -33.6906
30
- - Logits/rejected: -2.1139
31
- - Logits/chosen: -2.1185
32
 
33
  ## Model description
34
 
@@ -57,27 +57,15 @@ The following hyperparameters were used during training:
57
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
58
  - lr_scheduler_type: cosine
59
  - lr_scheduler_warmup_ratio: 0.1
60
- - num_epochs: 4
61
 
62
  ### Training results
63
 
64
- | Training Loss | Epoch | Step | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
65
- |:-------------:|:-----:|:----:|:-------------:|:---------------:|:------------:|:--------------:|:---------------:|:------------------:|:--------------:|:---------------:|:----------------:|
66
- | 0.4916 | 0.26 | 100 | -2.2338 | -2.2290 | -33.8981 | -37.4099 | 0.4981 | 0.5453 | 0.0409 | 0.0089 | 0.0320 |
67
- | 0.4782 | 0.52 | 200 | -2.2331 | -2.2283 | -33.7707 | -37.3050 | 0.4962 | 0.5718 | 0.0792 | 0.0157 | 0.0635 |
68
- | 0.4507 | 0.78 | 300 | -2.2302 | -2.2254 | -33.7484 | -37.2887 | 0.4956 | 0.5544 | 0.0859 | 0.0175 | 0.0684 |
69
- | 0.4139 | 1.04 | 400 | 0.4945 | 0.1190 | 0.0965 | 0.6063 | 0.0225 | -37.1951 | -33.6380 | -2.2093 | -2.2142 |
70
- | 0.3771 | 1.3 | 500 | 0.4937 | 0.1247 | 0.0982 | 0.5714 | 0.0264 | -37.1892 | -33.6190 | -2.1773 | -2.1820 |
71
- | 0.387 | 1.56 | 600 | 0.4881 | 0.1306 | 0.0798 | 0.5880 | 0.0508 | -37.2506 | -33.5992 | -2.1573 | -2.1621 |
72
- | 0.3565 | 1.82 | 700 | 0.4888 | 0.1084 | 0.0584 | 0.5710 | 0.0499 | -37.3218 | -33.6734 | -2.1444 | -2.1491 |
73
- | 0.3115 | 2.08 | 800 | 0.4843 | 0.1179 | 0.0481 | 0.6063 | 0.0698 | -37.3561 | -33.6414 | -2.1386 | -2.1433 |
74
- | 0.2849 | 2.34 | 900 | 0.4852 | 0.1036 | 0.0344 | 0.5826 | 0.0692 | -37.4021 | -33.6894 | -2.1241 | -2.1288 |
75
- | 0.336 | 2.6 | 1000 | 0.4836 | 0.1076 | 0.0285 | 0.6030 | 0.0791 | -37.4216 | -33.6757 | -2.1167 | -2.1214 |
76
- | 0.3426 | 2.86 | 1100 | 0.4845 | 0.0998 | 0.0255 | 0.6005 | 0.0743 | -37.4316 | -33.7020 | -2.1152 | -2.1199 |
77
- | 0.2883 | 3.12 | 1200 | 0.4838 | 0.1007 | 0.0221 | 0.6092 | 0.0785 | -37.4428 | -33.6989 | -2.1139 | -2.1186 |
78
- | 0.2779 | 3.38 | 1300 | 0.4829 | 0.1028 | 0.0215 | 0.6005 | 0.0813 | -37.4450 | -33.6919 | -2.1139 | -2.1185 |
79
- | 0.3113 | 3.64 | 1400 | 0.4838 | 0.1005 | 0.0222 | 0.5972 | 0.0784 | -37.4427 | -33.6995 | -2.1142 | -2.1189 |
80
- | 0.2932 | 3.9 | 1500 | 0.4838 | 0.1004 | 0.0226 | 0.6067 | 0.0779 | -37.4414 | -33.6997 | -2.1137 | -2.1184 |
81
 
82
 
83
  ### Framework versions
 
20
 
21
  This model is a fine-tuned version of [data/ap-gpt-j-6b-sft-qlora-04-08](https://huggingface.co/data/ap-gpt-j-6b-sft-qlora-04-08) on the hugodk-sch/aftonposten_title_prefs dataset.
22
  It achieves the following results on the evaluation set:
23
+ - Loss: 2.7902
24
+ - Rewards/chosen: -0.0060
25
+ - Rewards/rejected: -0.0171
26
+ - Rewards/accuracies: 0.5602
27
+ - Rewards/margins: 0.0111
28
+ - Logps/rejected: -37.5735
29
+ - Logps/chosen: -34.0545
30
+ - Logits/rejected: -2.2247
31
+ - Logits/chosen: -2.2295
32
 
33
  ## Model description
34
 
 
57
  - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
58
  - lr_scheduler_type: cosine
59
  - lr_scheduler_warmup_ratio: 0.1
60
+ - num_epochs: 1
61
 
62
  ### Training results
63
 
64
+ | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
65
+ |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|
66
+ | 2.59 | 0.26 | 100 | 2.8958 | 0.0029 | 0.0052 | 0.4730 | -0.0024 | -37.4993 | -34.0250 | -2.2305 | -2.2353 |
67
+ | 2.2795 | 0.52 | 200 | 2.8012 | -0.0060 | -0.0145 | 0.5278 | 0.0085 | -37.5651 | -34.0545 | -2.2290 | -2.2339 |
68
+ | 1.7902 | 0.78 | 300 | 2.7585 | -0.0030 | -0.0167 | 0.5748 | 0.0137 | -37.5724 | -34.0446 | -2.2245 | -2.2294 |
 
 
 
 
 
 
 
 
 
 
 
 
69
 
70
 
71
  ### Framework versions
adapter_config.json CHANGED
@@ -19,13 +19,13 @@
19
  "rank_pattern": {},
20
  "revision": null,
21
  "target_modules": [
22
- "q_proj",
23
- "gate_proj",
24
  "up_proj",
25
  "o_proj",
 
26
  "down_proj",
27
- "k_proj",
28
- "v_proj"
29
  ],
30
  "task_type": "CAUSAL_LM",
31
  "use_rslora": false
 
19
  "rank_pattern": {},
20
  "revision": null,
21
  "target_modules": [
22
+ "k_proj",
 
23
  "up_proj",
24
  "o_proj",
25
+ "v_proj",
26
  "down_proj",
27
+ "q_proj",
28
+ "gate_proj"
29
  ],
30
  "task_type": "CAUSAL_LM",
31
  "use_rslora": false
adapter_model.safetensors CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e4a47b71a074a3e66b22f474a623b02f20b5f4187dcf42a1804d77acb760541a
3
  size 176183216
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2803ff3a80b6c8d101cb2d67087ecd60ca0a5374db81896b8a2c14cc0e814a25
3
  size 176183216
all_results.json CHANGED
@@ -1,21 +1,21 @@
1
  {
2
- "epoch": 4.0,
3
- "eval_logits/chosen": -2.1185147762298584,
4
- "eval_logits/rejected": -2.1138522624969482,
5
- "eval_logps/chosen": -33.690555572509766,
6
- "eval_logps/rejected": -37.433555603027344,
7
- "eval_loss": 0.4836578071117401,
8
- "eval_rewards/accuracies": 0.6034052968025208,
9
- "eval_rewards/chosen": 0.10319855809211731,
10
- "eval_rewards/margins": 0.07827980816364288,
11
- "eval_rewards/rejected": 0.02491873875260353,
12
- "eval_runtime": 145.7738,
13
  "eval_samples": 343,
14
  "eval_samples_per_second": 2.353,
15
  "eval_steps_per_second": 0.295,
16
- "train_loss": 0.2714830153948301,
17
- "train_runtime": 10800.8138,
18
  "train_samples": 3079,
19
- "train_samples_per_second": 1.14,
20
- "train_steps_per_second": 0.143
21
  }
 
1
  {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -2.2295007705688477,
4
+ "eval_logits/rejected": -2.2246556282043457,
5
+ "eval_logps/chosen": -34.054466247558594,
6
+ "eval_logps/rejected": -37.57353973388672,
7
+ "eval_loss": 2.790188789367676,
8
+ "eval_rewards/accuracies": 0.560215950012207,
9
+ "eval_rewards/chosen": -0.0059735761024057865,
10
+ "eval_rewards/margins": 0.011101600714027882,
11
+ "eval_rewards/rejected": -0.017075177282094955,
12
+ "eval_runtime": 145.7539,
13
  "eval_samples": 343,
14
  "eval_samples_per_second": 2.353,
15
  "eval_steps_per_second": 0.295,
16
+ "train_loss": 2.3457992528940177,
17
+ "train_runtime": 3250.6949,
18
  "train_samples": 3079,
19
+ "train_samples_per_second": 0.947,
20
+ "train_steps_per_second": 0.118
21
  }
eval_results.json CHANGED
@@ -1,15 +1,15 @@
1
  {
2
- "epoch": 4.0,
3
- "eval_logits/chosen": -2.1185147762298584,
4
- "eval_logits/rejected": -2.1138522624969482,
5
- "eval_logps/chosen": -33.690555572509766,
6
- "eval_logps/rejected": -37.433555603027344,
7
- "eval_loss": 0.4836578071117401,
8
- "eval_rewards/accuracies": 0.6034052968025208,
9
- "eval_rewards/chosen": 0.10319855809211731,
10
- "eval_rewards/margins": 0.07827980816364288,
11
- "eval_rewards/rejected": 0.02491873875260353,
12
- "eval_runtime": 145.7738,
13
  "eval_samples": 343,
14
  "eval_samples_per_second": 2.353,
15
  "eval_steps_per_second": 0.295
 
1
  {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -2.2295007705688477,
4
+ "eval_logits/rejected": -2.2246556282043457,
5
+ "eval_logps/chosen": -34.054466247558594,
6
+ "eval_logps/rejected": -37.57353973388672,
7
+ "eval_loss": 2.790188789367676,
8
+ "eval_rewards/accuracies": 0.560215950012207,
9
+ "eval_rewards/chosen": -0.0059735761024057865,
10
+ "eval_rewards/margins": 0.011101600714027882,
11
+ "eval_rewards/rejected": -0.017075177282094955,
12
+ "eval_runtime": 145.7539,
13
  "eval_samples": 343,
14
  "eval_samples_per_second": 2.353,
15
  "eval_steps_per_second": 0.295
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 4.0,
3
- "train_loss": 0.2714830153948301,
4
- "train_runtime": 10800.8138,
5
  "train_samples": 3079,
6
- "train_samples_per_second": 1.14,
7
- "train_steps_per_second": 0.143
8
  }
 
1
  {
2
+ "epoch": 1.0,
3
+ "train_loss": 2.3457992528940177,
4
+ "train_runtime": 3250.6949,
5
  "train_samples": 3079,
6
+ "train_samples_per_second": 0.947,
7
+ "train_steps_per_second": 0.118
8
  }
trainer_state.json CHANGED
The diff for this file is too large to render. See raw diff
 
training_args.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:6796dc0cd2ca464da4b2c86fd637500a53c2e219bb1c345d39119ad06d09b09e
3
  size 4984
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:dc3e8127207cbcb9bf61a05f5206c35cfece8824544474dcbb3c0ba691e632f8
3
  size 4984