--- license: llama2 library_name: peft tags: - trl - dpo - generated_from_trainer base_model: meta-llama/Llama-2-7b-hf model-index: - name: llama_DPO_model_e2 results: [] --- # llama_DPO_model_e2 This model is a fine-tuned version of [meta-llama/Llama-2-7b-hf](https://huggingface.co/meta-llama/Llama-2-7b-hf) on an unknown dataset. It achieves the following results on the evaluation set: - Loss: 0.0037 - Rewards/chosen: 0.5612 - Rewards/rejected: -5.9460 - Rewards/accuracies: 1.0 - Rewards/margins: 6.5073 - Logps/rejected: -244.2698 - Logps/chosen: -155.0214 - Logits/rejected: -1.0632 - Logits/chosen: -0.8795 ## 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: 1 - eval_batch_size: 1 - seed: 42 - gradient_accumulation_steps: 8 - total_train_batch_size: 8 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: linear - num_epochs: 2 ### Training results | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| | 0.3693 | 0.1 | 25 | 0.1586 | 0.3906 | -1.4782 | 1.0 | 1.8688 | -199.5915 | -156.7276 | -1.0532 | -0.8639 | | 0.0442 | 0.2 | 50 | 0.0275 | 0.5577 | -3.3969 | 1.0 | 3.9546 | -218.7789 | -155.0573 | -1.0591 | -0.8709 | | 0.0153 | 0.3 | 75 | 0.0123 | 0.5805 | -4.3685 | 1.0 | 4.9490 | -228.4945 | -154.8291 | -1.0641 | -0.8765 | | 0.0098 | 0.4 | 100 | 0.0083 | 0.5880 | -4.8560 | 1.0 | 5.4440 | -233.3696 | -154.7535 | -1.0654 | -0.8801 | | 0.0072 | 0.5 | 125 | 0.0065 | 0.5779 | -5.1733 | 1.0 | 5.7513 | -236.5429 | -154.8546 | -1.0667 | -0.8808 | | 0.0056 | 0.6 | 150 | 0.0058 | 0.5669 | -5.3483 | 1.0 | 5.9152 | -238.2926 | -154.9651 | -1.0674 | -0.8815 | | 0.0059 | 0.7 | 175 | 0.0051 | 0.5733 | -5.4970 | 1.0 | 6.0704 | -239.7797 | -154.9004 | -1.0659 | -0.8820 | | 0.0065 | 0.79 | 200 | 0.0047 | 0.5713 | -5.6304 | 1.0 | 6.2017 | -241.1136 | -154.9210 | -1.0653 | -0.8803 | | 0.0044 | 0.89 | 225 | 0.0043 | 0.5689 | -5.7514 | 1.0 | 6.3203 | -242.3240 | -154.9452 | -1.0650 | -0.8816 | | 0.004 | 0.99 | 250 | 0.0041 | 0.5671 | -5.8118 | 1.0 | 6.3790 | -242.9280 | -154.9625 | -1.0644 | -0.8796 | | 0.0029 | 1.09 | 275 | 0.0040 | 0.5648 | -5.8589 | 1.0 | 6.4237 | -243.3990 | -154.9863 | -1.0633 | -0.8800 | | 0.0035 | 1.19 | 300 | 0.0038 | 0.5658 | -5.8892 | 1.0 | 6.4549 | -243.7013 | -154.9761 | -1.0630 | -0.8785 | | 0.0024 | 1.29 | 325 | 0.0039 | 0.5618 | -5.9044 | 1.0 | 6.4662 | -243.8535 | -155.0163 | -1.0628 | -0.8787 | | 0.0034 | 1.39 | 350 | 0.0038 | 0.5595 | -5.9136 | 1.0 | 6.4731 | -243.9456 | -155.0389 | -1.0632 | -0.8788 | | 0.0029 | 1.49 | 375 | 0.0038 | 0.5601 | -5.9328 | 1.0 | 6.4929 | -244.1375 | -155.0332 | -1.0634 | -0.8792 | | 0.003 | 1.59 | 400 | 0.0038 | 0.5605 | -5.9352 | 1.0 | 6.4957 | -244.1614 | -155.0284 | -1.0632 | -0.8793 | | 0.0021 | 1.69 | 425 | 0.0038 | 0.5593 | -5.9410 | 1.0 | 6.5003 | -244.2199 | -155.0412 | -1.0630 | -0.8792 | | 0.0036 | 1.79 | 450 | 0.0038 | 0.5605 | -5.9408 | 1.0 | 6.5013 | -244.2178 | -155.0292 | -1.0631 | -0.8794 | | 0.0031 | 1.89 | 475 | 0.0038 | 0.5567 | -5.9439 | 1.0 | 6.5006 | -244.2483 | -155.0666 | -1.0634 | -0.8782 | | 0.0032 | 1.99 | 500 | 0.0037 | 0.5612 | -5.9460 | 1.0 | 6.5073 | -244.2698 | -155.0214 | -1.0632 | -0.8795 | ### Framework versions - PEFT 0.8.2 - Transformers 4.38.1 - Pytorch 2.2.0+cu118 - Datasets 2.17.1 - Tokenizers 0.15.2