runner: add vector challenge cleanup#5
Open
chokevin wants to merge 2 commits into
Open
Conversation
Add the vectorsum_v2 benchmark path, standalone submission entrypoint, Rune dispatch wiring, and FSDP sweep knobs used by the kernel-challenge work. Make local Python tests self-contained by generating synthetic NCU CSV fixtures instead of depending on ignored runs/ artifacts, and ignore local .tmp scratch output. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Make the test target request the uv dev extra explicitly so ruff and pytest are stable after ordinary uv run commands. Use the public NVCR PyTorch base for GitHub Actions image builds while keeping the Dockerfile default private ACR base for the canonical Azure ACR build path. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.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
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Adds the
vectorsum_v2kernel-challenge track: a standalonesubmission.py, reusable vector-sum kernels, benchmark runner, CLI command, and Rune experiment/dispatch wiring. It also carries the FSDP sweep knobs used in the optimization work and makes the NCU parser tests self-contained with synthetic fixtures instead of ignoredruns/artifacts.Why
The kernel-challenge worktree had useful optimization code mixed with local scratch artifacts and environment assumptions. This cleans it onto a feature branch, restores generated profile consistency, ignores
.tmp/, and fixes Python test setup so a fresh checkout can run the gates afteruv sync --extra dev.Non-goals
Testing
uv sync --extra devuv run ruff format swordfish tests submission.pyuv run ruff check swordfish tests submission.pymake testuv run python -m swordfish.runner bench-vectorsum --backend torch --size 64 --dtype fp32 --repeats 1 --warmup 0 --iters 1 --device cpu --allow-cpu --arch-label a100 --out /tmp/sf-vectorsum-smoke.jsonuv run python -m swordfish.runner generate-rune-profiles --check --out infra/rune/profiles/swordfish-pack.yamlRisk
Low for local users: new vector-sum paths are additive and covered by CPU smoke tests. Cluster risk is limited to dispatch rendering for the new experiment and FSDP knobs; rollback is reverting this commit or using existing
gemm/liger-*experiment names.