--- license: apache-2.0 base_model: mistralai/Mistral-7B-Instruct-v0.2 tags: - trl - dpo - generated_from_trainer model-index: - name: mistralit2_1000_STEPS_1e8_rate_0.1_beta_DPO results: [] --- # mistralit2_1000_STEPS_1e8_rate_0.1_beta_DPO This model is a fine-tuned version of [mistralai/Mistral-7B-Instruct-v0.2](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.2) on an unknown dataset. It achieves the following results on the evaluation set: - Loss: 0.6920 - Rewards/chosen: -0.0058 - Rewards/rejected: -0.0082 - Rewards/accuracies: 0.5121 - Rewards/margins: 0.0024 - Logps/rejected: -28.6543 - Logps/chosen: -23.4436 - Logits/rejected: -2.8649 - Logits/chosen: -2.8652 ## 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-08 - train_batch_size: 4 - eval_batch_size: 1 - seed: 42 - 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_steps: 100 - training_steps: 1000 ### Training results | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| | 0.693 | 0.1 | 50 | 0.6928 | 0.0007 | -0.0000 | 0.4549 | 0.0007 | -28.5728 | -23.3792 | -2.8652 | -2.8654 | | 0.693 | 0.2 | 100 | 0.6920 | 0.0012 | -0.0011 | 0.4945 | 0.0023 | -28.5838 | -23.3741 | -2.8653 | -2.8655 | | 0.693 | 0.29 | 150 | 0.6923 | -0.0015 | -0.0033 | 0.4989 | 0.0018 | -28.6052 | -23.4006 | -2.8651 | -2.8653 | | 0.694 | 0.39 | 200 | 0.6923 | -0.0020 | -0.0037 | 0.4813 | 0.0017 | -28.6093 | -23.4058 | -2.8651 | -2.8653 | | 0.6916 | 0.49 | 250 | 0.6922 | -0.0026 | -0.0046 | 0.4879 | 0.0021 | -28.6189 | -23.4118 | -2.8651 | -2.8654 | | 0.6927 | 0.59 | 300 | 0.6920 | -0.0039 | -0.0063 | 0.5011 | 0.0023 | -28.6350 | -23.4253 | -2.8650 | -2.8653 | | 0.6941 | 0.68 | 350 | 0.6927 | -0.0048 | -0.0058 | 0.4659 | 0.0010 | -28.6304 | -23.4334 | -2.8650 | -2.8652 | | 0.6924 | 0.78 | 400 | 0.6922 | -0.0049 | -0.0068 | 0.4989 | 0.0019 | -28.6399 | -23.4345 | -2.8650 | -2.8653 | | 0.6919 | 0.88 | 450 | 0.6918 | -0.0056 | -0.0084 | 0.4857 | 0.0028 | -28.6562 | -23.4418 | -2.8650 | -2.8653 | | 0.6913 | 0.98 | 500 | 0.6913 | -0.0047 | -0.0085 | 0.5077 | 0.0038 | -28.6577 | -23.4328 | -2.8649 | -2.8652 | | 0.6914 | 1.07 | 550 | 0.6915 | -0.0034 | -0.0067 | 0.5143 | 0.0033 | -28.6398 | -23.4200 | -2.8650 | -2.8653 | | 0.6939 | 1.17 | 600 | 0.6922 | -0.0069 | -0.0089 | 0.5033 | 0.0020 | -28.6613 | -23.4550 | -2.8650 | -2.8652 | | 0.6917 | 1.27 | 650 | 0.6920 | -0.0056 | -0.0081 | 0.5231 | 0.0025 | -28.6535 | -23.4422 | -2.8650 | -2.8653 | | 0.6919 | 1.37 | 700 | 0.6921 | -0.0052 | -0.0074 | 0.5055 | 0.0021 | -28.6463 | -23.4383 | -2.8650 | -2.8653 | | 0.6929 | 1.46 | 750 | 0.6915 | -0.0044 | -0.0078 | 0.5363 | 0.0034 | -28.6506 | -23.4298 | -2.8650 | -2.8653 | | 0.6919 | 1.56 | 800 | 0.6922 | -0.0063 | -0.0083 | 0.5209 | 0.0020 | -28.6553 | -23.4489 | -2.8649 | -2.8652 | | 0.6925 | 1.66 | 850 | 0.6921 | -0.0058 | -0.0080 | 0.5121 | 0.0022 | -28.6528 | -23.4438 | -2.8649 | -2.8652 | | 0.6925 | 1.76 | 900 | 0.6920 | -0.0058 | -0.0082 | 0.5121 | 0.0024 | -28.6543 | -23.4436 | -2.8649 | -2.8652 | | 0.6939 | 1.86 | 950 | 0.6920 | -0.0058 | -0.0082 | 0.5121 | 0.0024 | -28.6543 | -23.4436 | -2.8649 | -2.8652 | | 0.6924 | 1.95 | 1000 | 0.6920 | -0.0058 | -0.0082 | 0.5121 | 0.0024 | -28.6543 | -23.4436 | -2.8649 | -2.8652 | ### Framework versions - Transformers 4.38.2 - Pytorch 2.0.0+cu117 - Datasets 2.18.0 - Tokenizers 0.15.2