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: 0.9913
- Rewards/chosen: -0.0569
- Rewards/rejected: -0.0657
- Rewards/accuracies: 0.5511
- Rewards/margins: 0.0088
- Logps/rejected: -44.0859
- Logps/chosen: -39.7278
- Logits/rejected: -1.5833
- Logits/chosen: -1.5872
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.9987 | 0.26 | 100 | -2.2313 | -2.2264 | -34.0499 | -37.5528 | 0.9998 | 0.5336 | -0.0002 | 0.0002 | -0.0004 |
| 0.9965 | 0.52 | 200 | -2.2252 | -2.2204 | -34.0618 | -37.5790 | 0.9996 | 0.5071 | -0.0003 | 0.0004 | -0.0006 |
| 0.9925 | 0.78 | 300 | -2.2214 | -2.2166 | -34.0836 | -37.6063 | 0.9996 | 0.5594 | -0.0005 | 0.0004 | -0.0009 |
| 0.986 | 1.04 | 400 | 0.9994 | -0.0014 | -0.0019 | 0.5212 | 0.0006 | -37.7105 | -34.1717 | -2.1877 | -2.1926 |
| 0.9781 | 1.3 | 500 | 0.9987 | -0.0031 | -0.0044 | 0.5855 | 0.0013 | -37.9551 | -34.3418 | -2.1137 | -2.1185 |
| 0.9774 | 1.56 | 600 | 0.9973 | -0.0073 | -0.0101 | 0.5743 | 0.0027 | -38.5228 | -34.7671 | -2.0162 | -2.0208 |
| 0.9688 | 1.82 | 700 | 0.9969 | -0.0143 | -0.0174 | 0.5482 | 0.0031 | -39.2598 | -35.4681 | -1.9235 | -1.9280 |
| 0.957 | 2.08 | 800 | 0.9954 | -0.0214 | -0.0260 | 0.5540 | 0.0046 | -40.1194 | -36.1733 | -1.8363 | -1.8407 |
| 0.9358 | 2.34 | 900 | 0.9939 | -0.0362 | -0.0423 | 0.5365 | 0.0061 | -41.7483 | -37.6532 | -1.6988 | -1.7029 |
| 0.9535 | 2.6 | 1000 | 0.9921 | -0.0511 | -0.0591 | 0.5453 | 0.0079 | -43.4237 | -39.1479 | -1.6143 | -1.6183 |
| 0.9616 | 2.86 | 1100 | 0.9916 | -0.0562 | -0.0646 | 0.5453 | 0.0084 | -43.9754 | -39.6505 | -1.5880 | -1.5920 |
| 0.9167 | 3.12 | 1200 | 0.9912 | -0.0563 | -0.0651 | 0.5482 | 0.0088 | -44.0289 | -39.6666 | -1.5851 | -1.5890 |
| 0.9033 | 3.38 | 1300 | 0.9913 | -0.0570 | -0.0657 | 0.5453 | 0.0087 | -44.0868 | -39.7316 | -1.5817 | -1.5856 |
| 0.9285 | 3.64 | 1400 | 0.9912 | -0.0569 | -0.0657 | 0.5395 | 0.0088 | -44.0852 | -39.7216 | -1.5825 | -1.5864 |
| 0.9196 | 3.9 | 1500 | 0.9913 | -0.0567 | -0.0655 | 0.5424 | 0.0088 | -44.0648 | -39.7075 | -1.5832 | -1.5871 |
Framework versions
- PEFT 0.8.2
- Transformers 4.37.2
- Pytorch 2.1.2+cu121
- Datasets 2.17.0
- Tokenizers 0.15.1