--- license: other base_model: deepseek-ai/deepseek-llm-7b-chat tags: - alignment-handbook - trl - dpo - generated_from_trainer - trl - dpo - generated_from_trainer datasets: - self-generate/ds_chat_original_cn_mining_oj_iter0-binarized - self-generate/ds_chat_original_cn_mining_sandbox_iter0-binarized - self-generate/ds_chat_original_cn_rl_oj_iter0-binarized model-index: - name: ds_chat_sppo_hard_new_iter0_2024-09-15-01.40 results: [] --- [Visualize in Weights & Biases](https://ml.byteintl.net/experiment/tracking/detail?Id=project_20240915_20321b8f&selectedTrial=run_20240915_973730b7) # ds_chat_sppo_hard_new_iter0_2024-09-15-01.40 This model is a fine-tuned version of [deepseek-ai/deepseek-llm-7b-chat](https://huggingface.co/deepseek-ai/deepseek-llm-7b-chat) on the self-generate/ds_chat_original_cn_mining_oj_iter0-binarized, the self-generate/ds_chat_original_cn_mining_sandbox_iter0-binarized and the self-generate/ds_chat_original_cn_rl_oj_iter0-binarized datasets. It achieves the following results on the evaluation set: - Loss: 0.4619 - Rewards/chosen: 0.0067 - Rewards/rejected: -0.0352 - Rewards/accuracies: 0.5921 - Rewards/margins: 0.0419 - Logps/rejected: -263.1805 - Logps/chosen: -252.2534 - Logits/rejected: 1.4436 - Logits/chosen: 1.3993 - Debug/policy Chosen Logits: 1.3993 - Debug/policy Rejected Logits: 1.4436 - Debug/policy Chosen Logps: -252.2534 - Debug/policy Rejected Logps: -263.1805 - Debug/reference Chosen Logps: -252.9185 - Debug/reference Rejected Logps: -259.6586 ## 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-07 - train_batch_size: 8 - eval_batch_size: 4 - seed: 42 - distributed_type: multi-GPU - num_devices: 8 - total_train_batch_size: 64 - total_eval_batch_size: 32 - optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 - lr_scheduler_type: linear - lr_scheduler_warmup_ratio: 0.1 - lr_scheduler_warmup_steps: 100 - num_epochs: 8.0 ### Training results | Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen | Debug/policy Chosen Logits | Debug/policy Rejected Logits | Debug/policy Chosen Logps | Debug/policy Rejected Logps | Debug/reference Chosen Logps | Debug/reference Rejected Logps | |:-------------:|:------:|:----:|:---------------:|:--------------:|:----------------:|:------------------:|:---------------:|:--------------:|:------------:|:---------------:|:-------------:|:--------------------------:|:----------------------------:|:-------------------------:|:---------------------------:|:----------------------------:|:------------------------------:| | 0.4973 | 0.3623 | 100 | 0.4977 | -0.0056 | -0.0071 | 0.5132 | 0.0014 | -260.3654 | -253.4812 | 1.6987 | 1.6372 | 1.6372 | 1.6987 | -253.4812 | -260.3654 | -252.9185 | -259.6586 | | 0.4917 | 0.7246 | 200 | 0.4919 | -0.0069 | -0.0126 | 0.5395 | 0.0058 | -260.9230 | -253.6065 | 1.6704 | 1.6087 | 1.6087 | 1.6704 | -253.6065 | -260.9230 | -252.9185 | -259.6586 | | 0.4837 | 1.0870 | 300 | 0.4862 | -0.0085 | -0.0167 | 0.5789 | 0.0082 | -261.3287 | -253.7711 | 1.6490 | 1.5905 | 1.5905 | 1.6490 | -253.7711 | -261.3287 | -252.9185 | -259.6586 | | 0.4821 | 1.4493 | 400 | 0.4822 | -0.0046 | -0.0173 | 0.5132 | 0.0127 | -261.3844 | -253.3754 | 1.6131 | 1.5560 | 1.5560 | 1.6131 | -253.3754 | -261.3844 | -252.9185 | -259.6586 | | 0.4724 | 1.8116 | 500 | 0.4773 | -0.0010 | -0.0181 | 0.4737 | 0.0171 | -261.4722 | -253.0200 | 1.5870 | 1.5328 | 1.5328 | 1.5870 | -253.0200 | -261.4722 | -252.9185 | -259.6586 | | 0.4677 | 2.1739 | 600 | 0.4750 | -0.0007 | -0.0218 | 0.5132 | 0.0212 | -261.8435 | -252.9872 | 1.5701 | 1.5167 | 1.5167 | 1.5701 | -252.9872 | -261.8435 | -252.9185 | -259.6586 | | 0.4625 | 2.5362 | 700 | 0.5077 | 0.0917 | 0.0741 | 0.6447 | 0.0176 | -252.2495 | -243.7507 | 1.5700 | 1.5133 | 1.5133 | 1.5700 | -243.7507 | -252.2495 | -252.9185 | -259.6586 | | 0.465 | 2.8986 | 800 | 0.4709 | -0.0024 | -0.0313 | 0.5658 | 0.0289 | -262.7887 | -253.1583 | 1.5298 | 1.4781 | 1.4781 | 1.5298 | -253.1583 | -262.7887 | -252.9185 | -259.6586 | | 0.4551 | 3.2609 | 900 | 0.4689 | -0.0039 | -0.0344 | 0.5658 | 0.0304 | -263.0977 | -253.3132 | 1.5177 | 1.4670 | 1.4670 | 1.5177 | -253.3132 | -263.0977 | -252.9185 | -259.6586 | | 0.4614 | 3.6232 | 1000 | 0.4687 | -0.0108 | -0.0450 | 0.5789 | 0.0342 | -264.1606 | -253.9997 | 1.5075 | 1.4592 | 1.4592 | 1.5075 | -253.9997 | -264.1606 | -252.9185 | -259.6586 | | 0.4579 | 3.9855 | 1100 | 0.4668 | 0.0012 | -0.0346 | 0.5789 | 0.0358 | -263.1156 | -252.7994 | 1.5016 | 1.4527 | 1.4527 | 1.5016 | -252.7994 | -263.1156 | -252.9185 | -259.6586 | | 0.4466 | 4.3478 | 1200 | 0.4663 | 0.0006 | -0.0344 | 0.5526 | 0.0349 | -263.0953 | -252.8606 | 1.4940 | 1.4448 | 1.4448 | 1.4940 | -252.8606 | -263.0953 | -252.9185 | -259.6586 | | 0.4696 | 4.7101 | 1300 | 0.4644 | 0.0027 | -0.0346 | 0.5921 | 0.0373 | -263.1194 | -252.6523 | 1.4687 | 1.4226 | 1.4226 | 1.4687 | -252.6523 | -263.1194 | -252.9185 | -259.6586 | | 0.4571 | 5.0725 | 1400 | 0.4643 | -0.0002 | -0.0394 | 0.5789 | 0.0392 | -263.5992 | -252.9413 | 1.4644 | 1.4177 | 1.4177 | 1.4644 | -252.9413 | -263.5992 | -252.9185 | -259.6586 | | 0.45 | 5.4348 | 1500 | 0.4637 | 0.0047 | -0.0343 | 0.5789 | 0.0390 | -263.0912 | -252.4461 | 1.4551 | 1.4102 | 1.4102 | 1.4551 | -252.4461 | -263.0912 | -252.9185 | -259.6586 | | 0.4561 | 5.7971 | 1600 | 0.4627 | 0.0063 | -0.0340 | 0.5921 | 0.0403 | -263.0588 | -252.2838 | 1.4579 | 1.4127 | 1.4127 | 1.4579 | -252.2838 | -263.0588 | -252.9185 | -259.6586 | | 0.4505 | 6.1594 | 1700 | 0.4616 | 0.0094 | -0.0319 | 0.6316 | 0.0413 | -262.8479 | -251.9740 | 1.4445 | 1.4000 | 1.4000 | 1.4445 | -251.9740 | -262.8479 | -252.9185 | -259.6586 | | 0.4563 | 6.5217 | 1800 | 0.4613 | 0.0084 | -0.0356 | 0.6053 | 0.0440 | -263.2198 | -252.0771 | 1.4420 | 1.3981 | 1.3981 | 1.4420 | -252.0771 | -263.2198 | -252.9185 | -259.6586 | | 0.4675 | 6.8841 | 1900 | 0.4616 | 0.0069 | -0.0366 | 0.6053 | 0.0435 | -263.3192 | -252.2319 | 1.4424 | 1.3959 | 1.3959 | 1.4424 | -252.2319 | -263.3192 | -252.9185 | -259.6586 | | 0.4502 | 7.2464 | 2000 | 0.4619 | 0.0071 | -0.0364 | 0.5789 | 0.0435 | -263.2976 | -252.2066 | 1.4432 | 1.3985 | 1.3985 | 1.4432 | -252.2066 | -263.2976 | -252.9185 | -259.6586 | | 0.4473 | 7.6087 | 2100 | 0.4623 | 0.0028 | -0.0403 | 0.5921 | 0.0431 | -263.6902 | -252.6375 | 1.4423 | 1.3964 | 1.3964 | 1.4423 | -252.6375 | -263.6902 | -252.9185 | -259.6586 | | 0.4508 | 7.9710 | 2200 | 0.4619 | 0.0067 | -0.0352 | 0.5921 | 0.0419 | -263.1805 | -252.2534 | 1.4436 | 1.3993 | 1.3993 | 1.4436 | -252.2534 | -263.1805 | -252.9185 | -259.6586 | ### Framework versions - Transformers 4.42.0 - Pytorch 2.3.0+cu121 - Datasets 2.14.6 - Tokenizers 0.19.1