--- 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: 0.0572 - Rewards/chosen: 0.4916 - Rewards/rejected: -2.5677 - Rewards/accuracies: 1.0 - Rewards/margins: 3.0592 - Logps/rejected: -210.4865 - Logps/chosen: -155.7183 - Logits/rejected: -1.0527 - Logits/chosen: -0.8611 ## 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: 1e-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 | |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| | 0.6588 | 0.1 | 25 | 0.6197 | 0.0430 | -0.1117 | 0.9633 | 0.1547 | -185.9265 | -160.2034 | -1.0522 | -0.8546 | | 0.5198 | 0.2 | 50 | 0.4923 | 0.1198 | -0.3424 | 0.9933 | 0.4622 | -188.2335 | -159.4357 | -1.0525 | -0.8554 | | 0.422 | 0.3 | 75 | 0.3707 | 0.2016 | -0.6277 | 1.0 | 0.8293 | -191.0862 | -158.6175 | -1.0532 | -0.8571 | | 0.3133 | 0.4 | 100 | 0.2775 | 0.2622 | -0.9287 | 1.0 | 1.1908 | -194.0961 | -158.0122 | -1.0529 | -0.8575 | | 0.2536 | 0.5 | 125 | 0.2077 | 0.3244 | -1.2160 | 1.0 | 1.5403 | -196.9694 | -157.3904 | -1.0527 | -0.8608 | | 0.181 | 0.6 | 150 | 0.1559 | 0.3746 | -1.5115 | 1.0 | 1.8860 | -199.9242 | -156.8883 | -1.0534 | -0.8595 | | 0.1457 | 0.7 | 175 | 0.1203 | 0.4136 | -1.7795 | 1.0 | 2.1931 | -202.6049 | -156.4983 | -1.0534 | -0.8620 | | 0.1072 | 0.79 | 200 | 0.0950 | 0.4439 | -2.0245 | 1.0 | 2.4684 | -205.0550 | -156.1949 | -1.0532 | -0.8613 | | 0.0921 | 0.89 | 225 | 0.0792 | 0.4625 | -2.2196 | 1.0 | 2.6821 | -207.0056 | -156.0085 | -1.0535 | -0.8604 | | 0.0732 | 0.99 | 250 | 0.0694 | 0.4721 | -2.3665 | 1.0 | 2.8387 | -208.4748 | -155.9124 | -1.0530 | -0.8609 | | 0.0703 | 1.09 | 275 | 0.0636 | 0.4762 | -2.4589 | 1.0 | 2.9351 | -209.3987 | -155.8720 | -1.0527 | -0.8600 | | 0.0554 | 1.19 | 300 | 0.0606 | 0.4841 | -2.5053 | 1.0 | 2.9894 | -209.8628 | -155.7928 | -1.0528 | -0.8614 | | 0.0532 | 1.29 | 325 | 0.0592 | 0.4869 | -2.5331 | 1.0 | 3.0200 | -210.1407 | -155.7649 | -1.0527 | -0.8606 | | 0.061 | 1.39 | 350 | 0.0580 | 0.4912 | -2.5550 | 1.0 | 3.0462 | -210.3595 | -155.7218 | -1.0525 | -0.8611 | | 0.0612 | 1.49 | 375 | 0.0573 | 0.4930 | -2.5633 | 1.0 | 3.0563 | -210.4424 | -155.7034 | -1.0527 | -0.8613 | | 0.0539 | 1.59 | 400 | 0.0576 | 0.4921 | -2.5602 | 1.0 | 3.0523 | -210.4118 | -155.7133 | -1.0529 | -0.8596 | | 0.0517 | 1.69 | 425 | 0.0570 | 0.4917 | -2.5691 | 1.0 | 3.0608 | -210.5005 | -155.7172 | -1.0529 | -0.8602 | | 0.0627 | 1.79 | 450 | 0.0570 | 0.4938 | -2.5669 | 1.0 | 3.0607 | -210.4783 | -155.6961 | -1.0532 | -0.8608 | | 0.0575 | 1.89 | 475 | 0.0574 | 0.4911 | -2.5664 | 1.0 | 3.0574 | -210.4731 | -155.7233 | -1.0528 | -0.8612 | | 0.0578 | 1.99 | 500 | 0.0572 | 0.4916 | -2.5677 | 1.0 | 3.0592 | -210.4865 | -155.7183 | -1.0527 | -0.8611 | ### Framework versions - PEFT 0.8.2 - Transformers 4.38.1 - Pytorch 2.2.0+cu118 - Datasets 2.17.1 - Tokenizers 0.15.2