Skip to content

Fix public CI to use hf-bench uv env#155

Draft
lirundong wants to merge 2 commits into
NVIDIA:tilegym_updatefrom
lirundong:davidli/fix/hf-bench-ci-uv
Draft

Fix public CI to use hf-bench uv env#155
lirundong wants to merge 2 commits into
NVIDIA:tilegym_updatefrom
lirundong:davidli/fix/hf-bench-ci-uv

Conversation

@lirundong

Copy link
Copy Markdown
Contributor

Why

The public CI jobs for the modularized HF benchmark were still running tests through the container system Python. That bypassed the hf-bench uv environment and caused missing dependency failures such as PyTorch not being available.

What

  • Build the hf-bench Docker image through the source target so the source tree and uv-managed project are available.
  • Run the ops/suites test command from modeling/transformers via uv run --locked --no-sync python -m pytest.
  • Run benchmark scripts via uv run --locked --no-sync bash ... so the benchmark subprocesses inherit the hf-bench uv environment.

Assumption

This workflow change assumes the GitLab-side hf-bench Dockerfile and project metadata changes are landed into the public branch before the workflow is expected to pass end to end.

Validation

  • Parsed .github/workflows/tilegym-ci.yml with Ruby YAML.
  • Confirmed the public diff only touches .github/workflows/tilegym-ci.yml.

@copy-pr-bot

copy-pr-bot Bot commented Jun 15, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

hannahli-nv added a commit that referenced this pull request Jun 15, 2026
Combines changes from Ocean MR !1690 and TileGym PR #155:

- modeling/transformers/Dockerfile: consolidate to a single 'source' stage
  that copies the full repo and runs uv sync --locked --extra dev.
  Drops the old 'wheel' stage and system-Python pip installs.
- modeling/transformers/pyproject.toml: replace pinned
  tilegym @ git+...@v1.3.0 with editable in-tree path (../../).
- modeling/transformers/uv.lock: regenerated for in-tree tilegym.
- .github/workflows/tilegym-ci.yml: update build target from 'wheel'
  to 'source'; run test-ops and test-benchmark via `uv run` so they
  use the uv venv (which has torch + in-tree tilegym), not system Python.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
hannahli-nv added a commit that referenced this pull request Jun 15, 2026
Combines changes from Ocean MR !1690 and TileGym PR #155:

- modeling/transformers/Dockerfile: consolidate to a single 'source' stage
  that copies the full repo and runs uv sync --locked --extra dev.
  Drops the old 'wheel' stage and system-Python pip installs.
- modeling/transformers/pyproject.toml: replace pinned
  tilegym @ git+...@v1.3.0 with editable in-tree path (../../).
- modeling/transformers/uv.lock: regenerated for in-tree tilegym.
- .github/workflows/tilegym-ci.yml: update build target from 'wheel'
  to 'source'; run test-ops and test-benchmark via `uv run` so they
  use the uv venv (which has torch + in-tree tilegym), not system Python.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant