Skip to content

Add ROCm mismatch bypass env flag #466

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
jiagaoxiang wants to merge 6 commits into from
Closed

Add ROCm mismatch bypass env flag #466

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
f17f473
Enhance ROCm version check: allow mismatch with NVTE_ALLOW_ROCM_MISMA…
jiagaoxiang Feb 28, 2026
e35e02c
Clarify ROCm mismatch bypass in assertion error.
jiagaoxiang Feb 28, 2026
be43d54
Update transformer_engine/common/__init__.py
jiagaoxiang Feb 28, 2026
bfe4c90
Log ROCm mismatch status when bypass is enabled.
jiagaoxiang Feb 28, 2026
50ab2e8
Only warn when mismatch bypass actually applies.
jiagaoxiang Feb 28, 2026
e117c9d
Update transformer_engine/common/__init__.py
jiagaoxiang Feb 28, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 16 additions & 1 deletion transformer_engine/common/__init__.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -382,7 +382,22 @@ def _load_core_library():
build_rocm_version = list(filter(lambda f: f.startswith("ROCM_VERSION:"), build_info))
if build_rocm_version:
build_rocm_version = build_rocm_version[0].split(":")[1].strip().split('.')[:2]
assert (rocm_version == build_rocm_version), f"ROCm {'.'.join(rocm_version)} is detected but the library is built for {'.'.join(build_rocm_version)}"
# Strict by default. Set NVTE_ALLOW_ROCM_MISMATCH=1 to bypass.
allow_rocm_mismatch = os.getenv("NVTE_ALLOW_ROCM_MISMATCH", "0").strip().lower() in ("1", "true", "yes")
mismatch_detected = rocm_version != build_rocm_version
if allow_rocm_mismatch and mismatch_detected:
_logger.warning(
"NVTE_ALLOW_ROCM_MISMATCH is enabled. Proceeding despite ROCm runtime/build "
"version mismatch (runtime=%s, build=%s).",
".".join(rocm_version),
".".join(build_rocm_version),
)
elif mismatch_detected:
raise RuntimeError(
f"ROCm {'.'.join(rocm_version)} is detected but the library is built for "
f"{'.'.join(build_rocm_version)}. Set NVTE_ALLOW_ROCM_MISMATCH=1 to bypass "
"this check at your own risk."
)
except FileNotFoundError:
pass

Loading