# openai/parameter-golf

Train the smallest LM you can that fits in 16MB. Best model wins!

Repository: https://github.com/openai/parameter-golf
Canonical: https://ross.abutalabs.com/products/parameter-golf
Language: Python
License: MIT
License Family: permissive
Last push: 2026-05-04T20:07:19+00:00

## Health v2 (maintenance only)
Score: 51/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 80, release rhythm 35, longevity 14
- inputs: {"age_days": 205, "days_push": 121, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 5174, forks 3277 (observed 2026-08-28T04:09:11.671381+00:00)

## What it is
An OpenAI-hosted challenge to train the best-performing language model that fits in a 16MB artifact and trains in under 10 minutes on 8xH100s, evaluated by bits-per-byte compression on FineWeb. It includes training code, leaderboard infrastructure, and compute grants to encourage creative architectures and compression techniques.

## Use cases
- train the smallest language model that fits in 16MB
- compete on a parameter-constrained LLM training benchmark
- experiment with model compression and quantization-aware training
- learn efficient LLM training techniques like parameter tying and depth recurrence
- benchmark novel architectures under fixed compute limits
- apply for compute credits to train small models

## When to choose
- you want a competitive benchmark for parameter-efficient LLM training
- you're exploring compression, QAT, or novel small-model architectures
- you want a structured challenge with a leaderboard and community

## When to avoid
- you need a production-ready language model or inference library
- you lack access to H100-class GPUs and don't want to request grants
- you need general-purpose training tooling rather than a challenge codebase

## Facets
- artifact type: learning-resource
- maturity: active
- function: llm-training, benchmarking, machine-learning, deep-learning
- domain: large-language-models, machine-learning, gpu-computing, tutorials
- platform: python
- tags: model-compression, competition, nanogpt, quantization, parameter-efficiency, challenge, gpu, linux

## Member repositories
- openai/parameter-golf (main) score 51

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:09:11.671381+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 2026-08-29T18:01:50.381739+00:00, confidence not recorded.
  - readme: https://github.com/openai/parameter-golf (fetched 2026-08-28T04:09:11.671381+00:00, sha 52220cd975e9)
- Data as of 2026-08-30T08:39:29.467469+00:00.