--- 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](https://huggingface.co./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: 0.6747 - Rewards/chosen: -0.2693 - Rewards/rejected: -0.4359 - Rewards/accuracies: 0.6009 - Rewards/margins: 0.1666 - Logps/rejected: -38.1393 - Logps/chosen: -34.4192 - Logits/rejected: -2.1170 - Logits/chosen: -2.1217 ## 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 | |:-------------:|:-----:|:----:|:-------------:|:---------------:|:------------:|:--------------:|:---------------:|:------------------:|:--------------:|:---------------:|:----------------:| | 0.6577 | 0.26 | 100 | -2.2338 | -2.2289 | -34.0400 | -37.5357 | 0.6949 | 0.5316 | -0.0038 | 0.0095 | -0.0133 | | 0.6156 | 0.52 | 200 | -2.2317 | -2.2268 | -34.0535 | -37.5578 | 0.6943 | 0.5191 | -0.0132 | 0.0156 | -0.0288 | | 0.5468 | 0.78 | 300 | -2.2277 | -2.2229 | -34.0684 | -37.5860 | 0.6903 | 0.5191 | -0.0237 | 0.0249 | -0.0486 | | 0.4243 | 1.04 | 400 | 0.6886 | -0.0963 | -0.1489 | 0.5540 | 0.0526 | -37.7293 | -34.1721 | -2.1933 | -2.1981 | | 0.353 | 1.3 | 500 | 0.6901 | -0.1994 | -0.2743 | 0.5569 | 0.0749 | -37.9085 | -34.3194 | -2.1884 | -2.1932 | | 0.3554 | 1.56 | 600 | 0.6763 | -0.1468 | -0.2572 | 0.5806 | 0.1103 | -37.8840 | -34.2443 | -2.1701 | -2.1749 | | 0.3333 | 1.82 | 700 | 0.6813 | -0.1817 | -0.2946 | 0.5743 | 0.1129 | -37.9375 | -34.2941 | -2.1549 | -2.1596 | | 0.2025 | 2.08 | 800 | 0.6800 | -0.2316 | -0.3667 | 0.5660 | 0.1351 | -38.0405 | -34.3655 | -2.1413 | -2.1461 | | 0.2153 | 2.34 | 900 | 0.6866 | -0.2482 | -0.3826 | 0.5835 | 0.1344 | -38.0632 | -34.3891 | -2.1292 | -2.1340 | | 0.2381 | 2.6 | 1000 | 0.6821 | -0.2624 | -0.4162 | 0.5864 | 0.1538 | -38.1112 | -34.4094 | -2.1207 | -2.1255 | | 0.1898 | 2.86 | 1100 | 0.6858 | -0.2673 | -0.4161 | 0.5831 | 0.1487 | -38.1110 | -34.4164 | -2.1188 | -2.1235 | | 0.2231 | 3.12 | 1200 | 0.6780 | -0.2626 | -0.4264 | 0.5889 | 0.1637 | -38.1257 | -34.4097 | -2.1175 | -2.1223 | | 0.164 | 3.38 | 1300 | 0.6834 | -0.2678 | -0.4194 | 0.5947 | 0.1516 | -38.1158 | -34.4172 | -2.1174 | -2.1221 | | 0.1562 | 3.64 | 1400 | 0.6753 | -0.2666 | -0.4361 | 0.5922 | 0.1696 | -38.1396 | -34.4154 | -2.1172 | -2.1219 | | 0.2163 | 3.9 | 1500 | 0.6831 | -0.2684 | -0.4218 | 0.5801 | 0.1534 | -38.1192 | -34.4180 | -2.1173 | -2.1220 | ### Framework versions - PEFT 0.10.0 - Transformers 4.39.0.dev0 - Pytorch 2.1.2+cu121 - Datasets 2.14.6 - Tokenizers 0.15.1