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: 0.4618
- Rewards/chosen: -0.7871
- Rewards/rejected: -6.6095
- Rewards/accuracies: 0.9375
- Rewards/margins: 5.8223
- Logps/rejected: -220.7533
- Logps/chosen: -104.2417
- Logits/rejected: -1.9929
- Logits/chosen: -2.4654
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5633 | 0.0 | 10 | 0.5184 | 0.0815 | -0.3179 | 1.0 | 0.3995 | -157.8381 | -95.5550 | -2.1002 | -2.5591 |
| 0.3936 | 0.0 | 20 | 0.2875 | 0.2689 | -1.1716 | 1.0 | 1.4404 | -166.3746 | -93.6817 | -2.0675 | -2.5604 |
| 0.2132 | 0.0 | 30 | 0.2000 | 0.0966 | -2.8012 | 0.9375 | 2.8977 | -182.6702 | -95.4047 | -2.0286 | -2.5520 |
| 0.034 | 0.0 | 40 | 0.4078 | -0.5324 | -5.2327 | 0.9375 | 4.7003 | -206.9856 | -101.6947 | -2.0143 | -2.4945 |
| 0.0665 | 0.0 | 50 | 0.4618 | -0.7871 | -6.6095 | 0.9375 | 5.8223 | -220.7533 | -104.2417 | -1.9929 | -2.4654 |
Framework versions
- PEFT 0.9.0
- Transformers 4.38.2
- Pytorch 2.0.1+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2