metadata

license: mit
datasets:
  - openai/summarize_from_feedback
  - openai/webgpt_comparisons
  - Dahoas/instruct-synthetic-prompt-responses
language:
  - en
metrics:
  - accuracy
tags:
  - reward-model
  - reward_model
  - RLHF

Reward model trained from human feedback

Reward model (RM) trained to predict which generated answer is better judged by a human, given a question.

RM are useful in these domain:

QA model evaluation
serves as reward score in RLHF

All models are train on these dataset with a same split seed across datasets (if validation split wasn't available)

webgpt_comparisons
summarize_from_feedback
synthetic-instruct-gptj-pairwise

Performance

Validation split accuracy

Model	WebGPT	Summary	SytheticGPT
electra-large-discriminator	59.30	68.66	99.85
deberta-v3-large	61.13	72.23	99.94
deberta-v3-base	59.07	66.84	99.85

Its likely SytheticGPT has somekind of surface pattern on the choosen-rejected pair which makes it trivial to differentiate between better the answer.