MLPerf
Project record
MLPerf is MLCommons' family of system performance benchmarks. Its suites cover training (time to train a model to a target quality), inference in datacenter, edge, mobile, and tiny settings, and also client PCs, storage, automotive, HPC training, and inference endpoints. Results are submitted in rounds, reviewed by the submitting organizations, and published together by MLCommons.1329
- Website: MLCommons benchmarks overview (external site: mlcommons.org)
- Repository: MLPerf Inference reference implementations (external site: github.com)
- Repository: MLPerf Training reference implementations (external site: github.com)
- Documentation: MLPerf Inference rules (external site: github.com)
- Documentation: MLPerf Training rules (external site: github.com)
- Paper: MLPerf Inference benchmark paper (arXiv 1911.02549) (external site: arxiv.org)
Availability and license
Overall availability
Reference implementations and rules are public on GitHub under Apache 2.0, and published results can be browsed on the MLCommons site. Submitting results requires a signed Contributor License Agreement, and use of results with the MLPerf trademark must follow MLCommons' results messaging guidelines.67819
Availability is separate from permission: read the license before using or redistributing.
Apache License 2.0 (MLPerf Inference reference implementations) (external site: raw.githubusercontent.com)6
Apache License 2.0 (MLPerf Training reference implementations) (external site: raw.githubusercontent.com)7
Apache License 2.0 (MLPerf policies) (external site: raw.githubusercontent.com)8
MLPerf is a registered trademark of MLCommons. The submission rules state that, after publication, code and results are public and free for use under the MLPerf Terms of Use. Benchmark datasets come from separate sources, and their terms were not reviewed for this record.19
Public materials checklist
| Item | Status | Notes and evidence |
|---|---|---|
| CodeIs the evaluation code published? | Public | Reference implementations for the Inference and Training suites are on GitHub under Apache 2.0; the Training README says they are starting points and not fully optimized.4567 |
| Tasks / dataAre the tasks or test data available? | Public | Each reference implementation includes instructions for downloading its dataset; the Training README points to MLCommons storage for downloads. Terms vary by dataset and were not reviewed individually.54 |
| MethodologyIs the method for scoring described? | Public | Rules for Inference and Training are published, along with general submission rules covering divisions, availability categories, peer review, and publication.239 |
| ReproducibilityAre instructions for reproducing results published? | Public | Reference implementations include Dockerfiles and run scripts, and submissions must include READMEs and software information so results can be replicated.59 |
| LimitationsAre known limitations documented? | Partial | The Training README warns that reference implementations are not intended for real performance measurement, and results pages note that published results are sometimes modified or invalidated, with a change log. No broader discussion of benchmark validity was reviewed.52 |
What it is useful for
Run and use notes
- At review, the Inference README's most recent round is MLPerf Inference v6.1 (submission deadline July 31, 2026), and the Inference: Datacenter page links to v6.1 results. The Training README lists MLPerf Training v6.1 (submission deadline October 16, 2026), while the Training results page shows v6.0 results.4253
Organization context
U.S. eligibility
Eligible · basis: U.S.-governed project
MLPerf is developed and published by MLCommons, which holds the MLPerf trademark and hosts the reference implementations and rules in its GitHub organization. MLCommons Association is listed by the IRS as a 501(c)(6) organization with a Dover, Delaware address.1410
Sources
This listing is not an endorsement, a safety assessment, or a federal approval.