# mlcommons/training

Reference implementations of MLPerf® training benchmarks

Repository: https://github.com/mlcommons/training
Canonical: https://ross.abutalabs.com/products/training
Homepage: https://mlcommons.org/en/groups/training
Language: Python
License: Apache-2.0
License Family: permissive
Topics: benchmark, machine-learning
Last push: 2026-08-17T21:06:01+00:00

## Health v2 (maintenance only)
Score: 67/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 98, release rhythm 8, longevity 100
- inputs: {"age_days": 3079, "days_push": 16, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1771, forks 591 (observed 2026-08-28T04:05:34.085506+00:00)

## What it is
Reference implementations of the MLPerf Training benchmark suite maintained by MLCommons, covering models from LLMs to recommendation and vision tasks. Each benchmark ships with model code, Dockerfiles, dataset download scripts, and timing harnesses to measure how fast systems train models to a target quality metric.

## Use cases
- benchmark gpu training performance for MLPerf submissions
- compare training speed of different hardware platforms
- reproduce MLPerf training benchmark results
- get starting-point implementations for training LLMs like Llama 3.1
- measure time to train models to a target quality metric
- set up dockerized training benchmark environments

## When to choose
- you are preparing an MLPerf Training submission
- you need standardized, reproducible training benchmarks across vendors
- you want reference model implementations with defined quality targets

## When to avoid
- you need fully optimized production training code
- you want real-world software or hardware performance numbers rather than benchmark baselines
- you need a general-purpose training framework

## Facets
- artifact type: dataset
- maturity: active
- function: benchmarking, machine-learning, llm-training
- domain: machine-learning, deep-learning, large-language-models, performance, gpu-computing
- platform: python
- tags: mlperf, training-benchmarks, reference-implementations, mlcommons, performance-measurement, linux, docker, gpu

## Member repositories
- mlcommons/training (main) score 67

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:34.085506+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 2026-08-30T03:26:01.008608+00:00, confidence not recorded.
  - readme: https://github.com/mlcommons/training (fetched 2026-08-28T04:05:34.085506+00:00, sha aac552cc395b)
  - homepage: https://mlcommons.org/en/groups/training (fetched 2026-08-29T11:04:23.340307+00:00, sha 73696d8e3c0f)
  - site_page: https://mlcommons.org/about-us (fetched 2026-08-29T11:04:23.353214+00:00, sha 4011a6b5f799)
  - site_page: https://mlcommons.org/about-us/leadership (fetched 2026-08-29T11:04:23.354934+00:00, sha a8932609aea7)
  - site_page: https://mlcommons.org/about-us/programs (fetched 2026-08-29T11:04:23.356949+00:00, sha a7f4c948fc45)
  - site_page: https://mlcommons.org/ailuminate/safety-faq (fetched 2026-08-29T11:04:23.348894+00:00, sha 504e79ed138f)
  - site_page: https://mlcommons.org/ailuminate/jailbreak-faq (fetched 2026-08-29T11:04:23.351390+00:00, sha 9bbfc00d0047)
- Data as of 2026-08-30T08:39:29.467469+00:00.