Fix expected test failures for predispatch export nested cond and out_dtype #108715
Conversation
…_dtype [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/108715
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 New Failure, 2 Unrelated FailuresAs of commit 0133141 with merge base e73ec92 ( NEW FAILURE - The following job has failed:
BROKEN TRUNK - The following jobs failed but were present on the merge base:👉 Rebase onto the `viable/strict` branch to avoid these failures
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
@pytorchbot merge |
Merge failedReason: Not merging any PRs at the moment because there is a merge blocking https://github.com/pytorch/pytorch/labels/ci:%20sev issue open at: Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
Merge failedReason: 1 jobs have failed, first few of them are: inductor / cuda12.1-py3.10-gcc9-sm86 / test (inductor, 1, 1, linux.g5.4xlarge.nvidia.gpu) Details for Dev Infra teamRaised by workflow job |
|
@pytorchbot merge -i |
Merge startedYour change will be merged while ignoring the following 3 checks: pull / linux-jammy-py3.8-gcc11 / test (docs_test, 1, 1, linux.2xlarge), pull / linux-docs / build-docs-python-false, inductor / cuda12.1-py3.10-gcc9-sm86 / test (inductor, 1, 1, linux.g5.4xlarge.nvidia.gpu) Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
3 participants
Stack from ghstack (oldest at bottom):
Before this PR, we use get_fake_value to get the fake_sub_args then call op(*fake_sub_args) to get the example value for out dtype.
This causes problem when the input proxy's op type is
get_attr, get_fake_value for aget_attrnode will actually look at the original param/buffer and return a real tensor instead of fake tensor. This is OK for export, since export's fake_mode allows non_fake_inputs see here. But it causes problem when nesting cond with out_dtype where cond will use torch.compile(full_graph=True) to inspect out_dtype and find the inputs to op are mixed FakeTensor and real tensor.This PR changes how we get the example values from proxies by directly looking at node.meta["example_value"]. This meta data is guaranteed to exist for all proxies during dynamo tracing so it's safe to use ( it's also used by get_fake_value to get fake tensors from args for general ops see here).
Test Plan:
existing tests + remove expected failure for a test.
cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @Xia-Weiwen @wenzhe-nrv @jiayisunx @chenyang78 @aakhundov @kadeng