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
ad52dca
•
1 Parent(s): 3cf4dea

Training in progress, step 400

Browse files
Files changed (8) hide show
  1. README.md +15 -27
  2. adapter_config.json +5 -5
  3. adapter_model.safetensors +1 -1
  4. all_results.json +16 -16
  5. eval_results.json +12 -12
  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: 0.6784
24
- - Rewards/chosen: -0.2292
25
- - Rewards/rejected: -0.3691
26
- - Rewards/accuracies: 0.5714
27
- - Rewards/margins: 0.1398
28
- - Logps/rejected: -38.1317
29
- - Logps/chosen: -34.4166
30
- - Logits/rejected: -2.1013
31
- - Logits/chosen: -2.1060
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
- | 0.6595 | 0.26 | 100 | -2.2336 | -2.2287 | -34.0063 | -37.5150 | 0.6900 | 0.5253 | 0.0170 | 0.0160 | 0.0010 |
67
- | 0.6189 | 0.52 | 200 | -2.2307 | -2.2259 | -34.0167 | -37.5344 | 0.6895 | 0.5341 | 0.0107 | 0.0213 | -0.0106 |
68
- | 0.5582 | 0.78 | 300 | -2.2269 | -2.2221 | -34.0353 | -37.5627 | 0.6872 | 0.5598 | -0.0004 | 0.0272 | -0.0276 |
69
- | 0.4405 | 1.04 | 400 | 0.6790 | 0.0114 | -0.0399 | 0.5714 | 0.0513 | -37.5832 | -34.0156 | -2.2017 | -2.2066 |
70
- | 0.3628 | 1.3 | 500 | 0.6817 | -0.0615 | -0.1311 | 0.5743 | 0.0696 | -37.7351 | -34.1370 | -2.1583 | -2.1631 |
71
- | 0.3765 | 1.56 | 600 | 0.6655 | -0.0823 | -0.2023 | 0.6038 | 0.1201 | -37.8538 | -34.1716 | -2.1403 | -2.1450 |
72
- | 0.3979 | 1.82 | 700 | 0.6831 | -0.1484 | -0.2490 | 0.5569 | 0.1006 | -37.9315 | -34.2819 | -2.1358 | -2.1405 |
73
- | 0.2521 | 2.08 | 800 | 0.6749 | -0.1425 | -0.2655 | 0.5889 | 0.1230 | -37.9591 | -34.2720 | -2.1299 | -2.1346 |
74
- | 0.209 | 2.34 | 900 | 0.6766 | -0.1824 | -0.3167 | 0.5889 | 0.1344 | -38.0445 | -34.3385 | -2.1150 | -2.1197 |
75
- | 0.2724 | 2.6 | 1000 | 0.6788 | -0.2111 | -0.3496 | 0.5860 | 0.1386 | -38.0994 | -34.3864 | -2.1061 | -2.1108 |
76
- | 0.2927 | 2.86 | 1100 | 0.6755 | -0.2205 | -0.3683 | 0.5947 | 0.1478 | -38.1305 | -34.4021 | -2.1021 | -2.1068 |
77
- | 0.2152 | 3.12 | 1200 | 0.6774 | -0.2277 | -0.3699 | 0.5772 | 0.1421 | -38.1331 | -34.4141 | -2.1017 | -2.1064 |
78
- | 0.1618 | 3.38 | 1300 | 0.6745 | -0.2219 | -0.3688 | 0.5889 | 0.1469 | -38.1312 | -34.4043 | -2.1011 | -2.1058 |
79
- | 0.216 | 3.64 | 1400 | 0.6788 | -0.2276 | -0.3654 | 0.5801 | 0.1377 | -38.1255 | -34.4140 | -2.1012 | -2.1059 |
80
- | 0.2086 | 3.9 | 1500 | 0.6755 | -0.2256 | -0.3710 | 0.5831 | 0.1454 | -38.1350 | -34.4106 | -2.1010 | -2.1057 |
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: 0.4947
24
+ - Rewards/chosen: 0.1656
25
+ - Rewards/rejected: 0.1437
26
+ - Rewards/accuracies: 0.5365
27
+ - Rewards/margins: 0.0219
28
+ - Logps/rejected: -37.2771
29
+ - Logps/chosen: -33.7585
30
+ - Logits/rejected: -2.2259
31
+ - Logits/chosen: -2.2307
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
+ | 0.4799 | 0.26 | 100 | 0.4987 | 0.0790 | 0.0734 | 0.5104 | 0.0056 | -37.3943 | -33.9029 | -2.2301 | -2.2349 |
67
+ | 0.4548 | 0.52 | 200 | 0.4956 | 0.1590 | 0.1392 | 0.5341 | 0.0198 | -37.2846 | -33.7696 | -2.2287 | -2.2335 |
68
+ | 0.41 | 0.78 | 300 | 0.4937 | 0.1639 | 0.1391 | 0.5361 | 0.0248 | -37.2848 | -33.7614 | -2.2261 | -2.2309 |
 
 
 
 
 
 
 
 
 
 
 
 
69
 
70
 
71
  ### Framework versions
adapter_config.json CHANGED
@@ -19,13 +19,13 @@
19
  "rank_pattern": {},
20
  "revision": null,
21
  "target_modules": [
22
- "v_proj",
23
- "q_proj",
24
- "down_proj",
25
- "o_proj",
26
  "gate_proj",
27
  "up_proj",
28
- "k_proj"
 
 
 
 
29
  ],
30
  "task_type": "CAUSAL_LM",
31
  "use_rslora": false
 
19
  "rank_pattern": {},
20
  "revision": null,
21
  "target_modules": [
 
 
 
 
22
  "gate_proj",
23
  "up_proj",
24
+ "k_proj",
25
+ "down_proj",
26
+ "o_proj",
27
+ "v_proj",
28
+ "q_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:c534c4a142155a731f22ad921f85e07479e4c79eeee27a41bba786b13f044f91
3
  size 176183216
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:b63c985a10b670eab11934ea00b012922b6d00399a977ccb4b3fb0946a95f2ef
3
  size 176183216
all_results.json CHANGED
@@ -1,21 +1,21 @@
1
  {
2
- "epoch": 4.0,
3
- "eval_logits/chosen": -2.106013536453247,
4
- "eval_logits/rejected": -2.10129714012146,
5
- "eval_logps/chosen": -34.416629791259766,
6
- "eval_logps/rejected": -38.131744384765625,
7
- "eval_loss": 0.67835533618927,
8
- "eval_rewards/accuracies": 0.5714285373687744,
9
- "eval_rewards/chosen": -0.22924686968326569,
10
- "eval_rewards/margins": 0.13982589542865753,
11
- "eval_rewards/rejected": -0.36907273530960083,
12
- "eval_runtime": 145.7883,
13
  "eval_samples": 343,
14
- "eval_samples_per_second": 2.353,
15
  "eval_steps_per_second": 0.295,
16
- "train_loss": 0.24285796547864938,
17
- "train_runtime": 10807.9192,
18
  "train_samples": 3079,
19
- "train_samples_per_second": 1.14,
20
- "train_steps_per_second": 0.142
21
  }
 
1
  {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -2.230694055557251,
4
+ "eval_logits/rejected": -2.225895643234253,
5
+ "eval_logps/chosen": -33.75852966308594,
6
+ "eval_logps/rejected": -37.27712631225586,
7
+ "eval_loss": 0.4946732223033905,
8
+ "eval_rewards/accuracies": 0.5365448594093323,
9
+ "eval_rewards/chosen": 0.16561271250247955,
10
+ "eval_rewards/margins": 0.02191758342087269,
11
+ "eval_rewards/rejected": 0.14369513094425201,
12
+ "eval_runtime": 145.7007,
13
  "eval_samples": 343,
14
+ "eval_samples_per_second": 2.354,
15
  "eval_steps_per_second": 0.295,
16
+ "train_loss": 0.46439293204963983,
17
+ "train_runtime": 3253.2697,
18
  "train_samples": 3079,
19
+ "train_samples_per_second": 0.946,
20
+ "train_steps_per_second": 0.118
21
  }
eval_results.json CHANGED
@@ -1,16 +1,16 @@
1
  {
2
- "epoch": 4.0,
3
- "eval_logits/chosen": -2.106013536453247,
4
- "eval_logits/rejected": -2.10129714012146,
5
- "eval_logps/chosen": -34.416629791259766,
6
- "eval_logps/rejected": -38.131744384765625,
7
- "eval_loss": 0.67835533618927,
8
- "eval_rewards/accuracies": 0.5714285373687744,
9
- "eval_rewards/chosen": -0.22924686968326569,
10
- "eval_rewards/margins": 0.13982589542865753,
11
- "eval_rewards/rejected": -0.36907273530960083,
12
- "eval_runtime": 145.7883,
13
  "eval_samples": 343,
14
- "eval_samples_per_second": 2.353,
15
  "eval_steps_per_second": 0.295
16
  }
 
1
  {
2
+ "epoch": 1.0,
3
+ "eval_logits/chosen": -2.230694055557251,
4
+ "eval_logits/rejected": -2.225895643234253,
5
+ "eval_logps/chosen": -33.75852966308594,
6
+ "eval_logps/rejected": -37.27712631225586,
7
+ "eval_loss": 0.4946732223033905,
8
+ "eval_rewards/accuracies": 0.5365448594093323,
9
+ "eval_rewards/chosen": 0.16561271250247955,
10
+ "eval_rewards/margins": 0.02191758342087269,
11
+ "eval_rewards/rejected": 0.14369513094425201,
12
+ "eval_runtime": 145.7007,
13
  "eval_samples": 343,
14
+ "eval_samples_per_second": 2.354,
15
  "eval_steps_per_second": 0.295
16
  }
train_results.json CHANGED
@@ -1,8 +1,8 @@
1
  {
2
- "epoch": 4.0,
3
- "train_loss": 0.24285796547864938,
4
- "train_runtime": 10807.9192,
5
  "train_samples": 3079,
6
- "train_samples_per_second": 1.14,
7
- "train_steps_per_second": 0.142
8
  }
 
1
  {
2
+ "epoch": 1.0,
3
+ "train_loss": 0.46439293204963983,
4
+ "train_runtime": 3253.2697,
5
  "train_samples": 3079,
6
+ "train_samples_per_second": 0.946,
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 sha256:5af63d4badccbe780519fbb021e605570213cb7c051d9d1499f6c7ffb571e837
3
  size 4984
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a61cf4921e25061d4673043facf2e8060cc927f0981742727fad132386961410
3
  size 4984