# openai/lm-human-preferences

Code for the paper Fine-Tuning Language Models from Human Preferences

Repository: https://github.com/openai/lm-human-preferences
Canonical: https://ross.abutalabs.com/products/lm-human-preferences
Homepage: https://openai.com/blog/fine-tuning-gpt-2/
Language: Python
License: MIT
License Family: permissive
Archived: true
Last push: 2023-07-25T16:48:57+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": 2545, "days_push": 1135, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases, archived
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1391, forks 171 (observed 2026-08-28T04:04:35.912253+00:00)

## What it is
OpenAI's research code for the paper 'Fine-Tuning Language Models from Human Preferences', implementing reward model training from human labels and RL fine-tuning of GPT-2 language models. It is an archived repository provided as-is with no updates expected.

## Use cases
- reproduce the fine-tuning from human preferences paper
- train a reward model from human preference labels
- fine-tune GPT-2 with reinforcement learning from human feedback
- study early RLHF implementations
- download released human preference labels for experiments

## When to choose
- you want to study or reproduce the original 2019 RLHF research code
- you need the exact reward modeling and PPO fine-tuning setup from the paper
- you want access to the released human preference label datasets

## When to avoid
- you need a modern, maintained RLHF library (use trl or similar)
- you need to fine-tune models larger than small GPT-2
- you need up-to-date TensorFlow or GPU support
- you want production-ready training pipelines

## Facets
- artifact type: library
- maturity: abandoned
- function: machine-learning, llm-training, reinforcement-learning
- domain: large-language-models, machine-learning, reinforcement-learning
- platform: python
- tags: rlhf, reward-model, human-preferences, gpt-2, research-code, tensorflow, archived, research, linux, gpu

## Member repositories
- openai/lm-human-preferences (main) score 10

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:35.912253+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 2026-08-30T04:39:37.029986+00:00, confidence not recorded.
  - readme: https://github.com/openai/lm-human-preferences (fetched 2026-08-28T04:04:35.912253+00:00, sha 5ff805a00ae4)
  - homepage: https://openai.com/blog/fine-tuning-gpt-2/ (fetched 2026-08-29T11:54:43.699701+00:00, sha 3657513dae3d)
  - site_page: https://openai.com/about (fetched 2026-08-29T11:54:43.708983+00:00, sha de4b627b33e3)
  - site_page: https://developers.openai.com/api/docs (fetched 2026-08-29T11:54:43.711000+00:00, sha d617c5215f2a)
  - site_page: https://developers.openai.com/ (fetched 2026-08-29T11:54:43.713277+00:00, sha 7d240c906018)
- Data as of 2026-08-30T08:39:29.467469+00:00.