Skip to content

Commit 6030f30

Browse files
justinchubyCopilot
andcommitted
Fix memcpy fix guidance: update ORT kernel registration, not lower opset
Address reviewer feedback: lowering opset is not the right approach. The correct fix for opset version mismatches is to update ORT kernel registration for the missing opset versions. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com> Signed-off-by: Justin Chu <justinchu@microsoft.com>
1 parent a721fb8 commit 6030f30

1 file changed

Lines changed: 3 additions & 2 deletions

File tree

  • .agents/skills/profiling-onnx-models

.agents/skills/profiling-onnx-models/SKILL.md

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -181,8 +181,9 @@ data transfer. Common culprits:
181181
- `ConstantOfShape` with unusual dtypes
182182

183183
**Fix:** See the `debugging-memcpy` skill for root-cause analysis and
184-
fix patterns. The most impactful fix is usually lowering opset version
185-
or replacing unsupported ops with CUDA-friendly alternatives.
184+
fix patterns. The correct fix is to update ORT kernel registration for
185+
the missing opset versions (e.g. Equal opset 19 CUDA gap). As a
186+
secondary option, replace unsupported ops with CUDA-friendly alternatives.
186187

187188
### 2. cuBLAS warmup spike
188189

0 commit comments

Comments
 (0)