metadata
license: apache-2.0
base_model: TheBloke/OpenHermes-2-Mistral-7B-GPTQ
tags:
- trl
- dpo
- generated_from_trainer
model-index:
- name: mistral-dpo
results: []
mistral-dpo
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: 0.0012
- Rewards/chosen: 23.3318
- Rewards/rejected: -6.7489
- Rewards/accuracies: 1.0
- Rewards/margins: 30.0806
- Logps/rejected: -87.3513
- Logps/chosen: -337.4500
- Logits/rejected: -1.2781
- Logits/chosen: -1.6769
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 |
|---|---|---|---|---|---|---|---|---|---|---|---|
| 0.5937 | 0.01 | 10 | 0.3601 | 0.7833 | -0.3607 | 0.9904 | 1.1441 | -23.4698 | -562.9344 | -1.1371 | -1.4401 |
| 0.0908 | 0.02 | 20 | 1.1245 | 8.8420 | -2.7352 | 0.9615 | 11.5772 | -47.2141 | -482.3473 | -1.1942 | -1.5504 |
| 0.0683 | 0.03 | 30 | 0.2541 | 17.6490 | -4.7403 | 0.9904 | 22.3893 | -67.2654 | -394.2778 | -1.2341 | -1.6426 |
| 0.0009 | 0.04 | 40 | 0.0015 | 22.5664 | -5.9863 | 1.0 | 28.5527 | -79.7251 | -345.1035 | -1.2763 | -1.6781 |
| 0.0003 | 0.05 | 50 | 0.0012 | 23.3318 | -6.7489 | 1.0 | 30.0806 | -87.3513 | -337.4500 | -1.2781 | -1.6769 |
Framework versions
- Transformers 4.35.2
- Pytorch 2.0.1+cu117
- Datasets 2.15.0
- Tokenizers 0.15.0