openhermes-mistral-dpo-gptq
This model is a fine-tuned version of TheBloke/OpenHermes-2-Mistral-7B-GPTQ on the None dataset. It achieves the following results on the evaluation set:
- Loss: 15.0898
- Rewards/chosen: -39.8716
- Rewards/rejected: -3.6932
- Rewards/accuracies: 0.1875
- Rewards/margins: -36.1783
- Logps/rejected: -252.6567
- Logps/chosen: -699.3276
- Logits/rejected: -1.1624
- Logits/chosen: -1.4009
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: 0.0002
- train_batch_size: 1
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 2
- training_steps: 50
- mixed_precision_training: Native AMP
Training results
Training Loss | Epoch | Step | Validation Loss | Rewards/chosen | Rewards/rejected | Rewards/accuracies | Rewards/margins | Logps/rejected | Logps/chosen | Logits/rejected | Logits/chosen |
---|---|---|---|---|---|---|---|---|---|---|---|
0.6897 | 0.01 | 10 | 0.6981 | -0.0237 | -0.0260 | 0.4375 | 0.0023 | -215.9846 | -300.8488 | -1.8495 | -2.0704 |
0.685 | 0.01 | 20 | 0.6805 | -0.0489 | -0.0934 | 0.5 | 0.0446 | -216.6584 | -301.1003 | -1.8466 | -2.0623 |
0.7153 | 0.01 | 30 | 0.6583 | 0.0084 | -0.1588 | 0.5 | 0.1672 | -217.3122 | -300.5275 | -1.8462 | -2.0615 |
0.7148 | 0.02 | 40 | 14.5562 | -38.4731 | -3.6173 | 0.25 | -34.8558 | -251.8970 | -685.3425 | -1.1606 | -1.4120 |
0.8198 | 0.03 | 50 | 15.0898 | -39.8716 | -3.6932 | 0.1875 | -36.1783 | -252.6567 | -699.3276 | -1.1624 | -1.4009 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.0.1+cu117
- Datasets 2.15.0
- Tokenizers 0.15.0
Model tree for yunjinchoi/openhermes-mistral-dpo-gptq
Base model
mistralai/Mistral-7B-v0.1
Finetuned
teknium/OpenHermes-2-Mistral-7B
Quantized
TheBloke/OpenHermes-2-Mistral-7B-GPTQ