lewtun's picture
lewtun HF staff
End of training
18a72f8 verified
metadata
license: other
base_model: lewtun/gemma-7b-sft-full-deita-10k-v0
tags:
  - alignment-handbook
  - trl
  - dpo
  - generated_from_trainer
  - trl
  - dpo
  - generated_from_trainer
datasets:
  - argilla/dpo-mix-7k
model-index:
  - name: gemma-7b-dpo-full-mix1-beta-0.4-epoch-3
    results: []

gemma-7b-dpo-full-mix1-beta-0.4-epoch-3

This model is a fine-tuned version of lewtun/gemma-7b-sft-full-deita-10k-v0 on the argilla/dpo-mix-7k dataset. It achieves the following results on the evaluation set:

  • Loss: 1.3044
  • Rewards/chosen: -6.0191
  • Rewards/rejected: -11.1290
  • Rewards/accuracies: 0.75
  • Rewards/margins: 5.1099
  • Logps/rejected: -479.3722
  • Logps/chosen: -468.4523
  • Logits/rejected: 98.8291
  • Logits/chosen: 104.8679

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: 2
  • eval_batch_size: 4
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 8
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 128
  • 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: 3.0

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.2569 1.9 100 1.0775 -6.7226 -12.3440 0.7396 5.6214 -482.4095 -470.2108 99.2330 105.2899

Framework versions

  • Transformers 4.39.0.dev0
  • Pytorch 2.1.2+cu121
  • Datasets 2.14.6
  • Tokenizers 0.15.1