[dynamo] Fix test_const_dict_variable_python_type to match upstream#3711
[dynamo] Fix test_const_dict_variable_python_type to match upstream#3711AKloniecki wants to merge 4 commits into
Conversation
Sync test with pytorch/pytorch#183500. The old test directly constructed ConstDictVariable outside of torch.compile, which requires an active InstructionTranslator that doesn't exist in a standalone test invocation. Replace with a torch.compile-based test that exercises the same dict/OrderedDict code paths. Fixes: #3697 Signed-off-by: Artur Kłoniecki <arturx.kloniecki@intel.com>
There was a problem hiding this comment.
Pull request overview
This PR updates an XPU Dynamo regression test to match upstream behavior by avoiding direct construction of ConstDictVariable outside of a torch.compile translation context (which can fail due to missing active InstructionTranslator).
Changes:
- Rewrites
test_const_dict_variable_python_typeto run the dict/OrderedDictpath undertorch.compile(..., backend="eager", fullgraph=True). - Replaces internal-variable assertions with an eager-vs-compiled output equivalence check.
Skill file(s) read: .github/skills/xpu-ops-pr-review/SKILL.md.
BBBela
left a comment
There was a problem hiding this comment.
It is 1 to 1 porting change from upstream to our local test copies, seems straightforward.
Thanks!
|
@copilot New UT failures detected in op_ut tests (266 total). Please check these new failures and analyze whether they are caused by the PR changes. Important: Do NOT update the plan in the PR description directly. Use reply comments to update the status. Job log: https://github.com/intel/torch-xpu-ops/actions/runs/26086830323 New failures:
... and 261 more failures. See full list in artifact. Instructions:
|
I checked run 26086830323 logs/artifacts and the 266 new Reproducer pytest commands from the failure artifact:
No PR-local fix is needed for |
Do not, ever, instruct your agent to push directly to my branches ever again. |
|
@copilot New UT failures detected in op_ut tests (266 total). Please check these new failures and analyze whether they are caused by the PR changes. Important: Do NOT update the plan in the PR description directly. Use reply comments to update the status. Job log: https://github.com/intel/torch-xpu-ops/actions/runs/26146533515 New failures:
... and 261 more failures. See full list in artifact. Instructions:
|
Checked run The PR-local code change remains only Reproducer pytest commands from the artifact:
No PR-local fix needed for |
@AKloniecki please rebase to the latest main branch to get clear PR timeline, without copilot auto comments |
Sync test with pytorch/pytorch#183500. The old test directly constructed ConstDictVariable outside of torch.compile, which requires an active InstructionTranslator that doesn't exist in a standalone test invocation. Replace with a torch.compile-based test that exercises the same dict/OrderedDict code paths.
Fixes: #3697