alvarobartt's picture
alvarobartt HF staff
End of training
7267e16 verified
metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-v0.1
tags:
  - trl
  - orpo
  - generated_from_trainer
model-index:
  - name: mistral-orpo-mix-b0.05-l1024-pl512-lr5e-7-cosine
    results: []

mistral-orpo-mix-b0.05-l1024-pl512-lr5e-7-cosine

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the None dataset. It achieves the following results on the evaluation set:

  • Loss: 0.8648
  • Rewards/chosen: -0.0405
  • Rewards/rejected: -0.0502
  • Rewards/accuracies: 0.6458
  • Rewards/margins: 0.0097
  • Logps/rejected: -1.0036
  • Logps/chosen: -0.8096
  • Logits/rejected: -2.9146
  • Logits/chosen: -2.9040
  • Nll Loss: 0.8392
  • Log Odds Ratio: -0.6215
  • Log Odds Chosen: 0.3802

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: 5e-07
  • train_batch_size: 8
  • eval_batch_size: 8
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • gradient_accumulation_steps: 2
  • total_train_batch_size: 64
  • total_eval_batch_size: 32
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: cosine
  • lr_scheduler_warmup_ratio: 0.1
  • lr_scheduler_warmup_steps: 100
  • num_epochs: 3

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen Nll Loss Log Odds Ratio Log Odds Chosen
0.9159 1.0 105 0.8794 -0.0421 -0.0499 0.6302 0.0078 -0.9975 -0.8413 -2.8931 -2.8875 0.8561 -0.6429 0.3024
0.8397 2.0 211 0.8612 -0.0404 -0.0495 0.6458 0.0092 -0.9902 -0.8071 -2.8882 -2.8794 0.8366 -0.6257 0.3555
0.7808 2.99 315 0.8648 -0.0405 -0.0502 0.6458 0.0097 -1.0036 -0.8096 -2.9146 -2.9040 0.8392 -0.6215 0.3802

Framework versions

  • Transformers 4.39.0
  • Pytorch 2.1.1+cu121
  • Datasets 2.16.1
  • Tokenizers 0.15.2