metadata
library_name: peft
tags:
- alignment-handbook
- trl
- dpo
- generated_from_trainer
base_model: NbAiLab/nb-gpt-j-6B-v2
datasets:
- hugodk-sch/aftonposten_title_prefs
model-index:
- name: aftonposten-6b-align-scan
results: []
aftonposten-6b-align-scan
This model is a fine-tuned version of data/ap-gpt-j-6b-sft-qlora-04-08 on the hugodk-sch/aftonposten_title_prefs dataset. It achieves the following results on the evaluation set:
- Loss: 0.6880
- Rewards/chosen: -0.2843
- Rewards/rejected: -0.4655
- Rewards/accuracies: 0.5835
- Rewards/margins: 0.1811
- Logps/rejected: -38.0338
- Logps/chosen: -34.3505
- Logits/rejected: -2.1374
- Logits/chosen: -2.1422
Model description
More information needed
Intended uses & limitations
More information needed
Training and evaluation data
More information needed
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 5e-06
- train_batch_size: 4
- eval_batch_size: 8
- seed: 42
- distributed_type: multi-GPU
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: cosine
- lr_scheduler_warmup_ratio: 0.1
- num_epochs: 4
Training results
| Training Loss | Epoch | Step | Logits/chosen | Logits/rejected | Logps/chosen | Logps/rejected | Validation Loss | Rewards/accuracies | Rewards/chosen | Rewards/margins | Rewards/rejected |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6464 | 0.26 | 100 | -2.2340 | -2.2291 | -34.0405 | -37.5500 | 0.6903 | 0.5685 | -0.0054 | 0.0246 | -0.0300 |
| 0.5931 | 0.52 | 200 | -2.2316 | -2.2267 | -34.0730 | -37.5769 | 0.6980 | 0.5158 | -0.0346 | 0.0196 | -0.0543 |
| 0.5301 | 0.78 | 300 | -2.2292 | -2.2243 | -34.0962 | -37.6000 | 0.6973 | 0.5390 | -0.0555 | 0.0195 | -0.0750 |
| 0.389 | 1.04 | 400 | 0.6933 | -0.1201 | -0.1849 | 0.5507 | 0.0649 | -37.7221 | -34.1680 | -2.1983 | -2.2032 |
| 0.322 | 1.3 | 500 | 0.7055 | -0.2815 | -0.3556 | 0.5515 | 0.0741 | -37.9118 | -34.3473 | -2.1969 | -2.2017 |
| 0.327 | 1.56 | 600 | 0.6703 | -0.1443 | -0.2944 | 0.5806 | 0.1500 | -37.8437 | -34.1949 | -2.1819 | -2.1867 |
| 0.3034 | 1.82 | 700 | 0.6868 | -0.1851 | -0.3175 | 0.5656 | 0.1323 | -37.8694 | -34.2402 | -2.1701 | -2.1749 |
| 0.1649 | 2.08 | 800 | 0.6812 | -0.2229 | -0.3850 | 0.5951 | 0.1621 | -37.9443 | -34.2822 | -2.1594 | -2.1642 |
| 0.1691 | 2.34 | 900 | 0.6881 | -0.2514 | -0.4183 | 0.5831 | 0.1669 | -37.9814 | -34.3138 | -2.1476 | -2.1524 |
| 0.1953 | 2.6 | 1000 | 0.6957 | -0.2986 | -0.4680 | 0.5918 | 0.1694 | -38.0366 | -34.3663 | -2.1400 | -2.1447 |
| 0.1463 | 2.86 | 1100 | 0.7010 | -0.3003 | -0.4559 | 0.5714 | 0.1555 | -38.0231 | -34.3682 | -2.1379 | -2.1427 |
| 0.1796 | 3.12 | 1200 | 0.6908 | -0.2876 | -0.4581 | 0.5748 | 0.1705 | -38.0257 | -34.3541 | -2.1376 | -2.1423 |
| 0.1264 | 3.38 | 1300 | 0.6911 | -0.2772 | -0.4526 | 0.5893 | 0.1755 | -38.0196 | -34.3425 | -2.1374 | -2.1422 |
| 0.1206 | 3.64 | 1400 | 0.6924 | -0.2868 | -0.4582 | 0.5918 | 0.1714 | -38.0257 | -34.3532 | -2.1371 | -2.1419 |
| 0.1645 | 3.9 | 1500 | 0.6943 | -0.2880 | -0.4573 | 0.5718 | 0.1693 | -38.0247 | -34.3545 | -2.1371 | -2.1419 |
Framework versions
- PEFT 0.10.0
- Transformers 4.39.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.15.1