[ShriyaRishab]

hello

New Round: v[X.Y]

Round chair: @
Prior round: v[]
Submission deadline:
Tracking issue: #

Use this checklist when the Training WG adopts a new round. For brand-new benchmarks added in the same round, also apply the new-benchmark label so that checklist is appended.

Key references:

• Submission rules & roadmap
• Training rules
• Compatibility table (Training)
• MLPerf Training benchmarks page

Confirm round scope with the Training WG

• Finalize benchmark list for v[X.Y]: carrying forward, new, and retired benchmarks
• Confirm submission deadline and ruleset version (e.g. 6.1.0)
• Confirm any benchmark-specific rule changes that carry into the new round

mlcommons/training — reference implementations

• Move retired benchmark folders under retired_benchmarks/
• Add a new round section to README.md with:
  • Submission deadline
  • Benchmark table (model, reference path, framework, dataset, parameter count)
  • Links to each active reference implementation
• For new benchmarks: merge reference code and follow the New Benchmark checklist (dataset upload, RCP logs, README, etc.)
• For carrying-forward benchmarks: confirm reference code still matches the closed-division rules for the new round

mlcommons/training_policies — rules document

Update training_rules.adoc:

• Update document date at the top of the file
• For carrying-forward benchmarks: set Latest version available to v[X.Y] in all active tables:
  • Benchmarks / datasets
  • Closed division models & targets
  • Hyperparameters
  • Quality / evaluation frequency
  • Minimum number of runs
• For new benchmarks: add rows to each table above plus any benchmark-specific appendix rules
• For retired benchmarks:
  • Remove from active tables
  • Move rules to Appendix: Deprecated benchmarks
  • Add a v[X.Y] appendix section if the round introduced round-specific exceptions (see v6.0 Specific Rules for an example)

mlcommons/logging — checkers & summarizer

• Add training_[X.Y].0/ compliance-checker configs under mlperf_logging/compliance_checker/
  • Copy forward common.yaml and per-benchmark closed/open YAMLs from the prior round; update for retired/new benchmarks
• Add or update RCP JSON files under mlperf_logging/rcp_checker/training_[X.Y].0/
• Update mlperf_logging/benchmark_meta.py:
  • Allowed benchmark list for the new ruleset
  • Required result-file counts per benchmark
• Add scripts/verify_for_v[X.Y]_training.sh (copy from prior round and update ruleset version)
• Update mlperf_logging/result_summarizer/config.yaml with the new ruleset column/benchmark list
• Tag a new logging release once changes are merged (submitters pin to this tag)

mlcommons/policies — compatibility table

Update MLPerf_Compatibility_Table.adoc:

• Add a column for Training v[X.Y]
• Mark supported benchmarks for the new round
• Mark all retired benchmarks as unsupported in the new column

Website & communications

Email marketing@mlcommons.org and ask them to update the Training benchmarks page:

• Carrying-forward benchmarks: update "Latest available version" to v[X.Y]
• Retired benchmarks: leave "Latest available version" unchanged
• New benchmarks: add a row for each new benchmark
• Sort rows by "Latest available version" (newest round additions at the top)

Optional but recommended before the first submission deadline:

• Publish benchmark blog posts on mlcommons.org for new or materially changed benchmarks
• Present round changes at a Training WG meeting

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅