metadata
library_name: peft
tags:
- trl
- dpo
- alignment-handbook
- generated_from_trainer
base_model: NbAiLab/nb-gpt-j-6B-v2
model-index:
- name: aftonposten-6b-align-scan
results: []
aftonposten-6b-align-scan
This model is a fine-tuned version of NbAiLab/nb-gpt-j-6B-v2 on the None dataset. It achieves the following results on the evaluation set:
- Loss: 0.9176
- Rewards/chosen: -0.0906
- Rewards/rejected: -0.1873
- Rewards/accuracies: 0.5627
- Rewards/margins: 0.0967
- Logps/rejected: -37.7248
- Logps/chosen: -34.1353
- Logits/rejected: -2.2000
- Logits/chosen: -2.2049
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.8913 | 0.26 | 100 | -2.2322 | -2.2273 | -34.0407 | -37.5405 | 0.9845 | 0.5195 | -0.0055 | 0.0159 | -0.0215 |
| 0.7293 | 0.52 | 200 | -2.2286 | -2.2238 | -34.0537 | -37.5811 | 0.9602 | 0.5714 | -0.0172 | 0.0408 | -0.0580 |
| 0.6144 | 0.78 | 300 | -2.2249 | -2.2201 | -34.0866 | -37.6032 | 0.9713 | 0.5282 | -0.0468 | 0.0310 | -0.0779 |
| 0.3632 | 1.04 | 400 | 0.9495 | -0.0909 | -0.1434 | 0.5602 | 0.0525 | -37.6760 | -34.1355 | -2.2076 | -2.2125 |
| 0.2994 | 1.3 | 500 | 0.9461 | -0.1647 | -0.2318 | 0.5540 | 0.0671 | -37.7742 | -34.2176 | -2.2162 | -2.2210 |
| 0.3408 | 1.56 | 600 | 0.9077 | -0.0675 | -0.1694 | 0.5868 | 0.1019 | -37.7048 | -34.1096 | -2.2017 | -2.2066 |
| 0.2796 | 1.82 | 700 | 0.9425 | -0.0929 | -0.1626 | 0.5569 | 0.0697 | -37.6973 | -34.1378 | -2.2012 | -2.2061 |
| 0.1052 | 2.08 | 800 | 0.9125 | -0.0848 | -0.1863 | 0.5926 | 0.1015 | -37.7236 | -34.1288 | -2.2003 | -2.2051 |
| 0.095 | 2.34 | 900 | 0.9005 | -0.0802 | -0.1942 | 0.5540 | 0.1140 | -37.7324 | -34.1237 | -2.2019 | -2.2067 |
| 0.123 | 2.6 | 1000 | 0.9194 | -0.0907 | -0.1876 | 0.5511 | 0.0969 | -37.7251 | -34.1353 | -2.1994 | -2.2043 |
| 0.0894 | 2.86 | 1100 | 0.9182 | -0.0915 | -0.1890 | 0.5336 | 0.0976 | -37.7267 | -34.1362 | -2.2001 | -2.2050 |
| 0.1086 | 3.12 | 1200 | 0.9023 | -0.0864 | -0.1976 | 0.5627 | 0.1112 | -37.7362 | -34.1306 | -2.2006 | -2.2054 |
| 0.0577 | 3.38 | 1300 | 0.9154 | -0.0922 | -0.1935 | 0.5598 | 0.1013 | -37.7317 | -34.1370 | -2.2002 | -2.2050 |
| 0.0375 | 3.64 | 1400 | 0.9233 | -0.0896 | -0.1810 | 0.5569 | 0.0914 | -37.7178 | -34.1342 | -2.2002 | -2.2050 |
| 0.0724 | 3.9 | 1500 | 0.9176 | -0.0906 | -0.1873 | 0.5627 | 0.0967 | -37.7248 | -34.1353 | -2.2000 | -2.2049 |
Framework versions
- PEFT 0.10.0
- Transformers 4.39.0.dev0
- Pytorch 2.1.2+cu121
- Datasets 2.14.6
- Tokenizers 0.15.1