amirabdullah19852020's picture
End of training
8b94c3a verified
metadata
license: mit
base_model: EleutherAI/gpt-neo-125m
tags:
  - trl
  - dpo
  - generated_from_trainer
model-index:
  - name: gpt-neo-125m_hh_reward
    results: []

gpt-neo-125m_hh_reward

This model is a fine-tuned version of EleutherAI/gpt-neo-125m on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.7503
  • Rewards/chosen: -4.2523
  • Rewards/rejected: -4.3731
  • Rewards/accuracies: 0.5625
  • Rewards/margins: 0.1208
  • Logps/rejected: -168.5040
  • Logps/chosen: -147.3926
  • Logits/rejected: -11.6528
  • Logits/chosen: -11.5062

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: 0.0001
  • train_batch_size: 16
  • 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: 150
  • training_steps: 4050

Training results

Training Loss Epoch Step Validation Loss Rewards/chosen Rewards/rejected Rewards/accuracies Rewards/margins Logps/rejected Logps/chosen Logits/rejected Logits/chosen
0.8022 0.2 2000 0.7737 -4.8718 -5.0523 0.5724 0.1805 -175.2956 -153.5872 -11.7730 -11.6673
0.7336 0.4 4000 0.7503 -4.2523 -4.3731 0.5625 0.1208 -168.5040 -147.3926 -11.6528 -11.5062

Framework versions

  • Transformers 4.38.2
  • Pytorch 2.0.1+cu117
  • Datasets 2.17.1
  • Tokenizers 0.15.2