Repository navigation
Add Maestro framework as a TorchBench user benchmark - #2715
Closed
x41lakazam wants to merge 8 commits into
Closed
x41lakazam wants to merge 8 commits into
x41lakazam wants to merge 8 commits into
Conversation
…indings pin pip install --no-build-isolation skips installing [build-system].requires, so hatchling (maestro's build backend) was never present and the build failed with BackendUnavailable. Also override the base image's PIP_CONSTRAINT pin on cuda-bindings/cuda-python for the maestro install, since cupti-python needs newer versions than that pin allows. Verified by building deploy/Dockerfile end to end against nvcr.io/nvidia/pytorch:25.06-py3 on swx-ucc-build01; pip install completes and `pip show maestro` confirms a clean install. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Brings userbenchmark/maestro/maestro up to date with the nccl-multi-test maestro engine, now made sync-ready on the refractor-for-pytorch branch: - core/orchestrator.py: generalize NCCL mem-pool registration to multiple axis process groups with explicit deregistration on destroy(); drop the is_root()-gated result collection so run_all_patterns() returns the full DataFrame on every rank. - core/axis.py, core/config_parser.py: per-axis environment variables via a new `envs` config key, threaded through to Axis/TorchAxis. - core/utils/distributed.py: add dist_breakpoint(); drop the standalone is_root() helper (inlined at call sites). - core/profilers/cupti.py: drop dead commented-out ActivityKind lines. - main.py, __main__.py: split run() into a thin CLI command plus _run_benchmark() returning the results DataFrame; switch to a relative package import. Also drops Megatron support to match the upstream engine, which removed it on the same branch: deletes blocks/megatron/, the megatron-core dependency, the Dockerfile's Megatron-LM install step, the megatron_moe_layer.yaml example, and the related doc mentions. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…overrides Pulled README.md and docs/doc.md content updates from nccl-multi-test/maestro (source of truth), and bumped pyproject.toml version/description to match. Re-applied this repo's layout-specific overrides on top: README.md keeps the docs/doc.md link and run.py usage snippet, pyproject.toml keeps the build-system/wheel packaging config needed for this repo's maestro/ layout. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
userbenchmark/maestro.Test plan
python run_benchmark.py maestroruns with an example config🤖 Generated with Claude Code