pytorch
898482f1 - [logging] log exceptions when provided (#111164)

Commit
1 year ago
[logging] log exceptions when provided (#111164) This PR will cause logging.exception() to also dump the exception and stacktrace. Copied from https://github.com/python/cpython/blob/74723e11109a320e628898817ab449b3dad9ee96/Lib/logging/__init__.py#L707-L711 repro: <details> ```python import torch import torch._inductor.config torch._inductor.config.triton.inject_relu_bug_TESTING_ONLY = "runtime_error" def fn(x, y): return (x @ y).relu() x, y = [torch.rand((16, 16), device='cuda') for _ in range (2)] torch.compile(fn)(x, y) ``` run with TORCHDYNAMO_REPRO_AFTER=aot TORCHDYNAMO_REPRO_LEVEL=4 </details> before: ``` ... [2023-10-12 14:18:52,902] torch._dynamo.debug_utils: [ERROR] While minifying the program in accuracy minification mode, ran into a runtime exception which is likely an unrelated issue. Skipping this graph. ``` now: ``` ... [2023-10-12 14:18:52,902] torch._dynamo.debug_utils: [ERROR] While minifying the program in accuracy minification mode, ran into a runtime exception which is likely an unrelated issue. Skipping this graph. Traceback (most recent call last): File "/data/users/dberard/scripts/relu_accuracy_issue.py", line 10, in <module> torch.compile(fn)(x, y) ... ``` Pull Request resolved: https://github.com/pytorch/pytorch/pull/111164 Approved by: https://github.com/eellison
Author
Committer
Parents
Loading