Laith Sakka comments

Results 19 comments of


                                            Laith Sakka

110 Dynamo test are failing with "Failed running call_function <built-in method sparse_coo_tensor of type object at 0xDEADBEEF".

Note that: 1. indices->device is CPU 2. indices->storage()->device is Meta?

110 Dynamo test are failing with "Failed running call_function <built-in method sparse_coo_tensor of type object at 0xDEADBEEF".

I looked at a different failure with in the set of failures, namely : "TestAssertCloseSparseBSC.test_mismatching_values_msg", # test_testing that is related to sparse_bsc_tensor I summarize my findings in https://docs.google.com/document/d/1EIb4qG88-SjVFn5TloLERliYdxIu2hwYoAA8skjOVfo/edit

110 Dynamo test are failing with "Failed running call_function <built-in method sparse_coo_tensor of type object at 0xDEADBEEF".

> I looked at a different failure with in the set of failures, namely : "TestAssertCloseSparseBSC.test_mismatching_values_msg", # test_testing that is related to sparse_bsc_tensor > > I summarize my findings in...

do not call has_pending_mutation with None input

landed as https://github.com/pytorch/pytorch/pull/125359

Fix dynamo issue "Failed running call_function <built-in method sparse_coo_tensor of type object at 0xDEADBEEF"

I wonder is shall i add sparce_coo_tensor to graph breaks for now in https://github.com/pytorch/pytorch/pull/120458

Investigate Error "f capturable=True, params and state_steps must be CUDA or XLA tensors" when nn module inlining enabled.

note: those only fails when dynamo is enabled on the test

Investigate Error "f capturable=True, params and state_steps must be CUDA or XLA tensors" when nn module inlining enabled.

I had offline chat with @mlazos he is already working on this or something very close to this, he said he will take over

Investigate Error "f capturable=True, params and state_steps must be CUDA or XLA tensors" when nn module inlining enabled.

@mlazos @anijain2305 this is fixed with recent Lazos changes

Investigate Error "f capturable=True, params and state_steps must be CUDA or XLA tensors" when nn module inlining enabled.

fixed by https://github.com/pytorch/pytorch/pull/125826