dmis-lab's picture
Update README.md
de02780 verified
|
raw
history blame
1.8 kB
metadata
base_model: Minbyul/selfbiorag-7b-wo-kqa_golden-iter-dpo-step3-filtered
tags:
  - alignment-handbook
  - trl
  - dpo
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: selfbiorag-7b-wo-kqa_golden-iter-dpo-step4-filtered
    results: []
language:
  - en

selfbiorag-7b-wo-kqa_golden-iter-dpo-step4-filtered

This model is a fine-tuned version of Minbyul/selfbiorag-7b-wo-kqa_golden-iter-dpo-step3-filtered on the HuggingFaceH4/ultrafeedback_binarized dataset. It achieves the following results on the evaluation set:

  • Loss: 0.6766
  • Rewards/chosen: -0.0828
  • Rewards/rejected: -0.1144
  • Rewards/accuracies: 0.6319
  • Rewards/margins: 0.0316
  • Logps/rejected: -98.9601
  • Logps/chosen: -79.1920
  • Logits/rejected: -1.2073
  • Logits/chosen: -1.1930

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

Framework versions

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