mlcommons / mlcommons/endpoints

Promote warmup to a first-class phase type with per-phase config

Open
#449 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
Python
Stars
21
Forks
28
Avg merge
3d 17h
Merged PRs (30d)
13

Description

Follow-up from PR #409 review (suggested by @arekay-nv in https://github.com/mlcommons/endpoints/pull/409#discussion_r3725383602).

Today warmup is a special-cased pre-phase: its knobs are scattered (settings.warmup block, settings.timeouts.warmup_drain_timeout_s) and phases are implicit. Proposal:

  • Explicit phase types with declared ordering/dependencies (warmup before performance, reporting after accuracy, ...).
  • Per-phase config blocks owning that phase's knobs (sample count, drain timeout, salt/seeds), replacing the flat per-phase drain fields in settings.timeouts.
  • The whole-run run_timeout_s watchdog (landed in #409) stays the only global deadline.

The Timeouts per-phase drain fields map 1:1 onto per-phase blocks when this lands.

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reviewing PR #409 and tracing the current settings.warmup block and settings.timeouts per-phase drain fields. Define explicit phase types, ordering and dependencies, then assign each phase its configuration while keeping run_timeout_s as the global watchdog; done means the warmup and existing phases use the new structure without the scattered fields.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
backend
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Mostly clear
Newbie friendliness
38/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.