# openai/summarize-from-feedback

Code for "Learning to summarize from human feedback"

Repository: https://github.com/openai/summarize-from-feedback
Canonical: https://ross.abutalabs.com/products/summarize-from-feedback
Language: Python
License: NOASSERTION
License Family: other
Archived: true
Last push: 2023-09-05T17:25:28+00:00

## Health v2 (maintenance only)
Score: 10/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 0, release rhythm 35, longevity 100
- inputs: {"age_days": 2191, "days_push": 1093, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived, no_license
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1062, forks 153 (observed 2026-08-28T04:03:26.166074+00:00)

## What it is
Research code and human feedback dataset from OpenAI's paper 'Learning to Summarize from human feedback', including a supervised baseline, a trained reward model, and an RL fine-tuned summarization policy. The repo is archived and provided as-is, with 64,832 summary comparisons on the TL;DR dataset plus evaluation data.

## Use cases
- reproduce rlhf summarization experiments from the openai paper
- download human feedback comparison data for summarization research
- run a trained reward model to score text summaries
- generate summaries with a ppo fine-tuned policy on a gpu
- study reward model evaluation on tldr and cnn/dm datasets

## When to choose
- you need the original human feedback dataset for rlhf research
- you want to reproduce or study the learning-to-summarize paper
- you need example code for reward models and ppo fine-tuning

## When to avoid
- you need maintained, up-to-date code or library support
- you want production summarization rather than research artifacts
- you cannot run python 3.7 on ubuntu with an nvidia gpu

## Facets
- artifact type: dataset
- maturity: abandoned
- function: machine-learning, deep-learning, nlp, llm-training, rag
- domain: machine-learning, large-language-models, artificial-intelligence
- platform: python
- tags: reinforcement-learning, human-feedback, rlhf, summarization, reward-model, research-code, tldr-dataset, archived, natural-language-processing, linux, gpu

## Member repositories
- openai/summarize-from-feedback (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:03:26.166074+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:56:45.864346+00:00, confidence not recorded.
  - readme: https://github.com/openai/summarize-from-feedback (fetched 2026-08-28T04:03:26.166074+00:00, sha c3751e187874)
- Data as of 2026-08-30T08:39:29.467469+00:00.