jdang's picture
jdang/openhermes-mistral-7b-dpo-gptq
3546419 verified
metadata
license: apache-2.0
base_model: TheBloke/OpenHermes-2-Mistral-7B-GPTQ
tags:
  - trl
  - dpo
  - generated_from_trainer
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: 0.6104
  • Rewards/chosen: -0.0458
  • Rewards/rejected: -0.4535
  • Rewards/accuracies: 0.6875
  • Rewards/margins: 0.4077
  • Logps/rejected: -390.3771
  • Logps/chosen: -149.5892
  • Logits/rejected: -1.3692
  • Logits/chosen: -1.4352

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
0.6865 0.01 10 0.6792 -0.0093 -0.0078 0.6875 -0.0015 -385.9200 -149.2238 -1.3698 -1.4189
0.6882 0.01 20 0.6660 -0.0137 -0.0526 0.625 0.0389 -386.3681 -149.2680 -1.3729 -1.4240
0.6391 0.01 30 0.6446 0.0000 -0.1131 0.625 0.1131 -386.9731 -149.1310 -1.3737 -1.4292
0.639 0.02 40 0.6271 -0.0337 -0.2758 0.6875 0.2421 -388.6000 -149.4686 -1.3729 -1.4342
0.6533 0.03 50 0.6104 -0.0458 -0.4535 0.6875 0.4077 -390.3771 -149.5892 -1.3692 -1.4352

Framework versions

  • Transformers 4.35.2
  • Pytorch 2.0.1+cu117
  • Datasets 2.16.1
  • Tokenizers 0.15.0