--- 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: 0.0739 - Rewards/chosen: 0.4632 - Rewards/rejected: -2.2899 - Rewards/accuracies: 1.0 - Rewards/margins: 2.7530 - Logps/rejected: -207.7081 - Logps/chosen: -156.0022 - Logits/rejected: -1.0521 - Logits/chosen: -0.8598 ## 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: 9e-07 - 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 | |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| | 0.6694 | 0.1 | 25 | 0.6365 | 0.0370 | -0.0813 | 0.9433 | 0.1183 | -185.6225 | -160.2637 | -1.0521 | -0.8545 | | 0.5526 | 0.2 | 50 | 0.5246 | 0.1015 | -0.2765 | 0.9967 | 0.3780 | -187.5744 | -159.6185 | -1.0524 | -0.8560 | | 0.4607 | 0.3 | 75 | 0.4173 | 0.1669 | -0.5106 | 1.0 | 0.6775 | -189.9152 | -158.9647 | -1.0530 | -0.8562 | | 0.3595 | 0.4 | 100 | 0.3251 | 0.2304 | -0.7635 | 1.0 | 0.9940 | -192.4449 | -158.3297 | -1.0530 | -0.8567 | | 0.297 | 0.5 | 125 | 0.2521 | 0.2883 | -1.0189 | 1.0 | 1.3072 | -194.9990 | -157.7509 | -1.0526 | -0.8573 | | 0.2217 | 0.6 | 150 | 0.1968 | 0.3313 | -1.2778 | 1.0 | 1.6090 | -197.5871 | -157.3212 | -1.0525 | -0.8576 | | 0.1832 | 0.7 | 175 | 0.1539 | 0.3750 | -1.5241 | 1.0 | 1.8991 | -200.0504 | -156.8834 | -1.0531 | -0.8606 | | 0.1374 | 0.79 | 200 | 0.1238 | 0.4055 | -1.7491 | 1.0 | 2.1546 | -202.3004 | -156.5787 | -1.0525 | -0.8614 | | 0.116 | 0.89 | 225 | 0.1027 | 0.4306 | -1.9426 | 1.0 | 2.3732 | -204.2353 | -156.3275 | -1.0526 | -0.8606 | | 0.095 | 0.99 | 250 | 0.0898 | 0.4405 | -2.0888 | 1.0 | 2.5293 | -205.6978 | -156.2289 | -1.0523 | -0.8603 | | 0.0921 | 1.09 | 275 | 0.0831 | 0.4465 | -2.1733 | 1.0 | 2.6198 | -206.5422 | -156.1685 | -1.0524 | -0.8593 | | 0.0734 | 1.19 | 300 | 0.0793 | 0.4520 | -2.2224 | 1.0 | 2.6744 | -207.0332 | -156.1135 | -1.0519 | -0.8627 | | 0.0711 | 1.29 | 325 | 0.0766 | 0.4558 | -2.2584 | 1.0 | 2.7142 | -207.3936 | -156.0763 | -1.0520 | -0.8592 | | 0.0806 | 1.39 | 350 | 0.0754 | 0.4630 | -2.2725 | 1.0 | 2.7355 | -207.5350 | -156.0041 | -1.0520 | -0.8599 | | 0.079 | 1.49 | 375 | 0.0748 | 0.4622 | -2.2779 | 1.0 | 2.7401 | -207.5887 | -156.0115 | -1.0522 | -0.8602 | | 0.0711 | 1.59 | 400 | 0.0746 | 0.4615 | -2.2817 | 1.0 | 2.7432 | -207.6269 | -156.0192 | -1.0519 | -0.8603 | | 0.0689 | 1.69 | 425 | 0.0744 | 0.4624 | -2.2862 | 1.0 | 2.7486 | -207.6718 | -156.0103 | -1.0522 | -0.8594 | | 0.0809 | 1.79 | 450 | 0.0742 | 0.4631 | -2.2887 | 1.0 | 2.7518 | -207.6965 | -156.0032 | -1.0517 | -0.8610 | | 0.0759 | 1.89 | 475 | 0.0740 | 0.4629 | -2.2902 | 1.0 | 2.7531 | -207.7117 | -156.0047 | -1.0517 | -0.8594 | | 0.0758 | 1.99 | 500 | 0.0739 | 0.4632 | -2.2899 | 1.0 | 2.7530 | -207.7081 | -156.0022 | -1.0521 | -0.8598 | ### Framework versions - PEFT 0.8.2 - Transformers 4.38.1 - Pytorch 2.2.0+cu118 - Datasets 2.17.1 - Tokenizers 0.15.2