Skip to content

fix(eval): route Modal runs to the requested export - #244

Open
rylinjames wants to merge 1 commit into
mainfrom
fix/modal-eval-export-dir
Open

fix(eval): route Modal runs to the requested export#244
rylinjames wants to merge 1 commit into
mainfrom
fix/modal-eval-export-dir

Conversation

@rylinjames

Copy link
Copy Markdown
Collaborator

Summary

  • map tether eval --runtime modal <export_dir> to a concrete /onnx_out/<subdir> Modal volume export instead of falling back to the script default
  • pass --onnx-subdir on every Modal suite invocation
  • document the Phase 1 limitation that local export files are not uploaded automatically

Validation

  • PYTHONPATH=src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/python -m pytest tests/test_eval_modal_runner.py tests/test_eval_cli.py -p no:cacheprovider
  • PYTHONPATH=src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/ruff check src/tether/eval/modal_runner.py tests/test_eval_modal_runner.py

🤖 Generated with Claude Code

The Modal LIBERO wrapper accepted export_dir but never passed it to the wrapped script, so tether eval --runtime modal could evaluate the default smolvla_libero_monolithic reference export while reporting results for a user's path.

Map the requested export directory to the Modal ONNX volume subdir and pass --onnx-subdir on every suite invocation. Local paths map by basename; /onnx_out paths preserve their relative subdir. If the volume subdir is missing, Modal now fails loudly instead of silently using the baked-in reference export.

Tests: PYTHONPATH=src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/python -m pytest tests/test_eval_modal_runner.py tests/test_eval_cli.py -p no:cacheprovider

Lint: PYTHONPATH=src /Users/romirjain/Desktop/building\ projects/fastcrest/tether/.venv/bin/ruff check src/tether/eval/modal_runner.py tests/test_eval_modal_runner.py

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

1 participant