metadata
license: apache-2.0
library_name: peft
tags:
- trl
- dpo
- generated_from_trainer
base_model: TheBloke/OpenHermes-2-Mistral-7B-GPTQ
model-index:
- name: openhermes-mistral-dpo-gptq
results: []
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: 0.6864
- Rewards/chosen: -0.0176
- Rewards/rejected: -0.0077
- Rewards/accuracies: 0.3125
- Rewards/margins: -0.0098
- Logps/rejected: -168.4619
- Logps/chosen: -127.7056
- Logits/rejected: -2.4072
- Logits/chosen: -2.3077
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.6768 | 0.005 | 10 | 0.6965 | 0.0206 | -0.0070 | 0.5625 | 0.0276 | -168.4548 | -127.3242 | -2.4001 | -2.3165 |
| 0.7062 | 0.01 | 20 | 0.7131 | 0.0043 | 0.0421 | 0.25 | -0.0378 | -167.9636 | -127.4874 | -2.4078 | -2.3156 |
| 0.7526 | 0.015 | 30 | 0.7101 | -0.0048 | 0.0478 | 0.125 | -0.0526 | -167.9063 | -127.5779 | -2.4088 | -2.3121 |
| 0.6946 | 0.02 | 40 | 0.7003 | -0.0144 | 0.0180 | 0.1875 | -0.0324 | -168.2044 | -127.6736 | -2.4087 | -2.3099 |
| 0.6728 | 0.025 | 50 | 0.6864 | -0.0176 | -0.0077 | 0.3125 | -0.0098 | -168.4619 | -127.7056 | -2.4072 | -2.3077 |
Framework versions
- PEFT 0.11.1
- Transformers 4.41.1
- Pytorch 2.0.1+cu117
- Datasets 2.19.1
- Tokenizers 0.19.1