Minbyul's picture
End of training
34618a4 verified
metadata
license: apache-2.0
base_model: mistralai/Mistral-7B-v0.1
tags:
  - alignment-handbook
  - trl
  - dpo
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
datasets:
  - HuggingFaceH4/ultrafeedback_binarized
model-index:
  - name: mistral-7b-dpo-full-wo-kqa_golden-ep3
    results: []

mistral-7b-dpo-full-wo-kqa_golden-ep3

This model is a fine-tuned version of mistralai/Mistral-7B-v0.1 on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

  • Loss: 0.2182
  • Rewards/chosen: -1.5510
  • Rewards/rejected: -6.3512
  • Rewards/accuracies: 0.875
  • Rewards/margins: 4.8002
  • Logps/rejected: -1537.8291
  • Logps/chosen: -720.7828
  • Logits/rejected: -2.8834
  • Logits/chosen: -3.1394

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
  • num_epochs: 1

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.1645 0.53 100 0.2503 -1.6026 -5.5026 0.8313 3.9001 -1452.9772 -725.9427 -2.7934 -3.0544

Framework versions

  • Transformers 4.39.0.dev0
  • Pytorch 2.1.2
  • Datasets 2.14.6
  • Tokenizers 0.15.2