fix(profiling): tolerate groot dep-install failure

groot's only policy-specific dependency is flash-attn, which has no
prebuilt wheel for torch 2.10 and requires nvcc to build from source.
The CI image is based on nvidia/cuda:12.4.1-base, which ships the
CUDA runtime but not the compiler toolkit, so the source build fails
with `/usr/local/cuda/bin/nvcc: No such file or directory`. The
repo's own pyproject.toml already carries a TODO acknowledging this:
gr00t needs bespoke flash-attn install steps.

Treat this as an environmental limitation rather than a regression:
dep-install failures for groot are logged via `::warning::` and skip
the policy without failing the job. Dep-install failures for any
other policy remain fatal, so real regressions still surface.

Made-with: Cursor
This commit is contained in:
Pepijn
2026-04-16 21:15:14 +02:00
parent 1842100402
commit a6dd28e8b4
+18 -2
View File
@@ -161,6 +161,18 @@ jobs:
esac
}
# Policies whose dep-install may fail due to environment constraints
# (e.g. groot requires compiling flash-attn, which needs nvcc; the CI
# image only ships the CUDA runtime). Install failures for these are
# logged as warnings and do not fail the job. See the TODO next to
# `lerobot[groot]` in pyproject.toml.
is_install_failure_tolerated() {
case "$1" in
groot) return 0 ;;
*) return 1 ;;
esac
}
overall_status=0
for raw_policy in "${policies_to_run[@]}"; do
policy="$(echo "${raw_policy}" | xargs)"
@@ -183,8 +195,12 @@ jobs:
# build isolation for flash-attn specifically.
sync_cmd+=(--no-build-isolation-package flash-attn)
if ! "${sync_cmd[@]}"; then
echo "Dependency install failed for ${policy}; skipping." >&2
overall_status=1
if is_install_failure_tolerated "${policy}"; then
echo "::warning::Dependency install failed for ${policy} (known-fragile); skipping."
else
echo "Dependency install failed for ${policy}; skipping." >&2
overall_status=1
fi
echo "::endgroup::"
continue
fi