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: 0.0132
- Rewards/chosen: -1.4792
- Rewards/rejected: -8.5855
- Rewards/accuracies: 1.0
- Rewards/margins: 7.1064
- Logps/rejected: -319.5252
- Logps/chosen: -138.4254
- Logits/rejected: -2.3872
- Logits/chosen: -2.5369
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5575 | 0.0 | 10 | 0.4017 | 0.0150 | -0.6143 | 1.0 | 0.6293 | -239.8125 | -123.4837 | -2.4102 | -2.6084 |
| 0.3781 | 0.0 | 20 | 0.1298 | -0.2390 | -2.2414 | 1.0 | 2.0025 | -256.0842 | -126.0231 | -2.3786 | -2.6120 |
| 0.219 | 0.0 | 30 | 0.0410 | -0.5640 | -4.3638 | 1.0 | 3.7998 | -277.3080 | -129.2739 | -2.3879 | -2.5872 |
| 0.038 | 0.0 | 40 | 0.0168 | -1.2083 | -7.3369 | 1.0 | 6.1286 | -307.0389 | -135.7168 | -2.3962 | -2.5566 |
| 0.0669 | 0.0 | 50 | 0.0132 | -1.4792 | -8.5855 | 1.0 | 7.1064 | -319.5252 | -138.4254 | -2.3872 | -2.5369 |
Framework versions
- PEFT 0.9.0
- Transformers 4.38.2
- Pytorch 2.0.1+cu117
- Datasets 2.18.0
- Tokenizers 0.15.2