--- license: apache-2.0 base_model: mistralai/Mistral-7B-Instruct-v0.1 tags: - trl - dpo - generated_from_trainer model-index: - name: v1_1000_STEPS_1e6_rate_03_beta_DPO2 results: [] --- # v1_1000_STEPS_1e6_rate_03_beta_DPO2 This model is a fine-tuned version of [mistralai/Mistral-7B-Instruct-v0.1](https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1) on an unknown dataset. It achieves the following results on the evaluation set: - Loss: 0.8487 - Rewards/chosen: -2.2151 - Rewards/rejected: -3.0240 - Rewards/accuracies: 0.5758 - Rewards/margins: 0.8089 - Logps/rejected: -26.9594 - Logps/chosen: -22.6366 - Logits/rejected: -3.2869 - Logits/chosen: -3.2870 ## 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: 2 - eval_batch_size: 1 - seed: 42 - gradient_accumulation_steps: 2 - total_train_batch_size: 4 - 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 | |:-------------:|:-----:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:| | 0.7016 | 0.05 | 50 | 0.6686 | -0.1010 | -0.1780 | 0.5516 | 0.0771 | -17.4730 | -15.5896 | -3.3830 | -3.3830 | | 0.6727 | 0.1 | 100 | 0.7732 | -1.2186 | -1.6229 | 0.5275 | 0.4043 | -22.2891 | -19.3150 | -3.3350 | -3.3352 | | 1.2098 | 0.15 | 150 | 0.9205 | -1.6685 | -2.0242 | 0.5209 | 0.3558 | -23.6270 | -20.8147 | -3.3998 | -3.4000 | | 0.8607 | 0.2 | 200 | 0.9312 | -1.7362 | -2.0915 | 0.5099 | 0.3553 | -23.8513 | -21.0405 | -3.3324 | -3.3326 | | 0.896 | 0.24 | 250 | 0.9765 | -1.8658 | -2.0921 | 0.5011 | 0.2263 | -23.8533 | -21.4723 | -3.2214 | -3.2215 | | 0.9783 | 0.29 | 300 | 0.9234 | -1.9658 | -2.3835 | 0.5165 | 0.4177 | -24.8244 | -21.8057 | -3.3158 | -3.3160 | | 1.0592 | 0.34 | 350 | 0.9509 | -3.1300 | -3.4037 | 0.5033 | 0.2738 | -28.2253 | -25.6863 | -3.2697 | -3.2698 | | 1.0391 | 0.39 | 400 | 0.9067 | -2.4562 | -2.8182 | 0.5231 | 0.3619 | -26.2735 | -23.4405 | -3.3616 | -3.3617 | | 0.9409 | 0.44 | 450 | 0.9081 | -2.8095 | -3.1865 | 0.5231 | 0.3771 | -27.5014 | -24.6179 | -3.3324 | -3.3325 | | 0.8139 | 0.49 | 500 | 0.9131 | -2.8071 | -3.2564 | 0.5560 | 0.4493 | -27.7343 | -24.6100 | -3.3362 | -3.3363 | | 0.8732 | 0.54 | 550 | 0.8745 | -2.3409 | -3.0357 | 0.5516 | 0.6948 | -26.9986 | -23.0562 | -3.3124 | -3.3125 | | 0.8179 | 0.59 | 600 | 0.8632 | -2.1460 | -2.9478 | 0.5692 | 0.8018 | -26.7055 | -22.4063 | -3.3039 | -3.3040 | | 0.825 | 0.64 | 650 | 0.8769 | -1.9605 | -2.7326 | 0.5626 | 0.7721 | -25.9882 | -21.7879 | -3.3006 | -3.3007 | | 0.7539 | 0.68 | 700 | 0.8600 | -2.1758 | -2.9531 | 0.5714 | 0.7773 | -26.7232 | -22.5059 | -3.2794 | -3.2795 | | 0.7835 | 0.73 | 750 | 0.8551 | -2.2525 | -3.0394 | 0.5692 | 0.7868 | -27.0107 | -22.7614 | -3.2905 | -3.2906 | | 0.925 | 0.78 | 800 | 0.8479 | -2.2131 | -3.0235 | 0.5736 | 0.8105 | -26.9579 | -22.6299 | -3.2902 | -3.2903 | | 1.0166 | 0.83 | 850 | 0.8493 | -2.2090 | -3.0157 | 0.5780 | 0.8067 | -26.9319 | -22.6164 | -3.2872 | -3.2873 | | 1.0711 | 0.88 | 900 | 0.8480 | -2.2126 | -3.0221 | 0.5758 | 0.8095 | -26.9532 | -22.6283 | -3.2869 | -3.2870 | | 0.9928 | 0.93 | 950 | 0.8487 | -2.2161 | -3.0255 | 0.5802 | 0.8094 | -26.9646 | -22.6400 | -3.2869 | -3.2870 | | 0.6707 | 0.98 | 1000 | 0.8487 | -2.2151 | -3.0240 | 0.5758 | 0.8089 | -26.9594 | -22.6366 | -3.2869 | -3.2870 | ### Framework versions - Transformers 4.39.2 - Pytorch 2.0.0+cu117 - Datasets 2.18.0 - Tokenizers 0.15.2