[quant] Fix dimension for output of batchnorm 1d

[jerryzh168]

Stack from ghstack:

• [quant] Fix dimension for output of batchnorm 1d #59264 [quant] Fix dimension for output of batchnorm 1d

Summary:
fixes: #59200
Previously batchnorm 1d did unsqueeze twice but only squeeze once before return when the dimension
for input Tensor is 2, this PR adds an extra squeeze

Test Plan:

Reviewers:

Subscribers:

Tasks:

Tags:

Differential Revision: D28810597

[facebook-github-bot]

💊 CI failures summary and remediations

As of commit 6070d79 (more details on the Dr. CI page):

---

• 3/3 failures possibly* introduced in this PR
  • 1/3 non-scanned failure(s)

---

🕵️ 1 new failure recognized by patterns

The following CI failures do not appear to be due to upstream breakages:

pytorch_linux_bionic_py3_6_clang9_noarch_test (1/1)

Step: "Run tests" (full log | diagnosis details | 🔁 rerun)

Jun 08 02:33:26 test_udf_remote_message_delay...yUniqueId(created_on=0, local_id=0) to be created.

Jun 08 02:32:39 frame #13: c10::ThreadPool::main_loop(unsigned long) + 0x17a (0x7ff3ef6d0e5a in /opt/conda/lib/python3.6/site-packages/torch/lib/libc10.so)
Jun 08 02:32:39 frame #14: <unknown function> + 0xc819d (0x7ff3ef5e619d in /opt/conda/lib/libstdc++.so.6)
Jun 08 02:32:39 frame #15: <unknown function> + 0x76db (0x7ff40d2786db in /lib/x86_64-linux-gnu/libpthread.so.0)
Jun 08 02:32:39 frame #16: clone + 0x3f (0x7ff40cfa171f in /lib/x86_64-linux-gnu/libc.so.6)
Jun 08 02:32:39 
Jun 08 02:32:40 ok (5.322s)
Jun 08 02:32:56   test_rpc_builtin_timeout (__main__.FaultyFaultyAgentRpcTestWithSpawn) ... ok (16.412s)
Jun 08 02:33:07   test_rpc_script_timeout (__main__.FaultyFaultyAgentRpcTestWithSpawn) ... ok (10.695s)
Jun 08 02:33:12   test_rref_to_here_timeout (__main__.FaultyFaultyAgentRpcTestWithSpawn) ... ok (5.174s)
Jun 08 02:33:21   test_udf_remote_message_delay_timeout (__main__.FaultyFaultyAgentRpcTestWithSpawn) ... ok (8.795s)
Jun 08 02:33:26   test_udf_remote_message_delay_timeout_to_self (__main__.FaultyFaultyAgentRpcTestWithSpawn) ... [E request_callback_no_python.cpp:552] Received error while processing request type 261: falseINTERNAL ASSERT FAILED at "/var/lib/jenkins/workspace/torch/csrc/distributed/rpc/rref_context.cpp":387, please report a bug to PyTorch. Expected OwnerRRef with id GloballyUniqueId(created_on=0, local_id=0) to be created.
Jun 08 02:33:26 Exception raised from getOwnerRRef at /var/lib/jenkins/workspace/torch/csrc/distributed/rpc/rref_context.cpp:387 (most recent call first):
Jun 08 02:33:26 frame #0: c10::Error::Error(c10::SourceLocation, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> >) + 0x7d (0x7fa7fa6df08d in /opt/conda/lib/python3.6/site-packages/torch/lib/libc10.so)
Jun 08 02:33:26 frame #1: c10::detail::torchCheckFail(char const*, char const*, unsigned int, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) + 0xde (0x7fa7fa6dd7ae in /opt/conda/lib/python3.6/site-packages/torch/lib/libc10.so)
Jun 08 02:33:26 frame #2: c10::detail::torchInternalAssertFail(char const*, char const*, unsigned int, char const*, std::__cxx11::basic_string<char, std::char_traits<char>, std::allocator<char> > const&) + 0x3b (0x7fa7fa6dd9fb in /opt/conda/lib/python3.6/site-packages/torch/lib/libc10.so)
Jun 08 02:33:26 frame #3: torch::distributed::rpc::RRefContext::getOwnerRRef(torch::distributed::rpc::GloballyUniqueId const&, bool) + 0x664 (0x7fa7fee5c094 in /opt/conda/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so)
Jun 08 02:33:26 frame #4: torch::distributed::rpc::RequestCallbackNoPython::assignOwnerRRef(torch::distributed::rpc::GloballyUniqueId const&, torch::distributed::rpc::GloballyUniqueId const&, c10::intrusive_ptr<c10::ivalue::Future, c10::detail::intrusive_target_default_null_type<c10::ivalue::Future> >) const + 0x59 (0x7fa7fee44b09 in /opt/conda/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so)
Jun 08 02:33:26 frame #5: torch::distributed::rpc::RequestCallbackImpl::processPythonRemoteCall(torch::distributed::rpc::RpcCommandBase&, std::vector<c10::Stream, std::allocator<c10::Stream> >) const + 0xa7 (0x7fa808145e87 in /opt/conda/lib/python3.6/site-packages/torch/lib/libtorch_python.so)
Jun 08 02:33:26 frame #6: torch::distributed::rpc::RequestCallbackNoPython::processRpc(torch::distributed::rpc::RpcCommandBase&, torch::distributed::rpc::MessageType const&, std::vector<c10::Stream, std::allocator<c10::Stream> >) const + 0x1d7 (0x7fa7fee43407 in /opt/conda/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so)
Jun 08 02:33:26 frame #7: torch::distributed::rpc::RequestCallbackImpl::processRpcWithErrors(torch::distributed::rpc::RpcCommandBase&, torch::distributed::rpc::MessageType const&, std::vector<c10::Stream, std::allocator<c10::Stream> >) const + 0x41 (0x7fa808146e31 in /opt/conda/lib/python3.6/site-packages/torch/lib/libtorch_python.so)
Jun 08 02:33:26 frame #8: <unknown function> + 0x4538e28 (0x7fa7fee4be28 in /opt/conda/lib/python3.6/site-packages/torch/lib/libtorch_cpu.so)

---

1 failure not recognized by patterns:

Job	Step	Action
Windows CI (pytorch-win-vs2019-cpu-py3) / render_test_results	Unknown	🔁 rerun

---

This comment was automatically generated by Dr. CI (expand for details).

Follow this link to opt-out of these comments for your Pull Requests.

Please report bugs/suggestions to the (internal) Dr. CI Users group.

Click here to manually regenerate this comment.

[jerryzh168]

@jerryzh168 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[pmeier]

@jerryzh168 You probably need to disable the 1d case in

pytorch/test/quantization/core/test_quantized_op.py

Line 2104 in 6d45d7a

def test_batch_norm_relu(self):

and

pytorch/test/quantization/core/test_quantized_op.py

Line 2155 in 6d45d7a

def test_batch_norm(self):

They only pass now because there is a bug in the underlying comparison mechanism. Try put a assert qx.shape == qy.shape in there and see for yourself. After #59067 has landed, the tests will (correctly) fail if not fixed.

[jerryzh168]

@jerryzh168 You probably need to disable the 1d case in

pytorch/test/quantization/core/test_quantized_op.py

Line 2104 in 6d45d7a

def test_batch_norm_relu(self):

and

pytorch/test/quantization/core/test_quantized_op.py

Line 2155 in 6d45d7a

def test_batch_norm(self):

They only pass now because there is a bug in the underlying comparison mechanism. Try put a assert qx.shape == qy.shape in there and see for yourself. After #59067 has landed, the tests will (correctly) fail if not fixed.

makes sense, this PR would fix the issue though, if we land this PR before yours it would not fail I think?

[pmeier]

If I'm not mistaken you only need to remove the 2 from

pytorch/test/quantization/core/test_quantized_op.py

Line 2106 in 6d45d7a

max_sides = (2, 3, 4, 5)

as well as the ==2 check here

pytorch/test/quantization/core/test_quantized_op.py

Line 2130 in 6d45d7a

if len(X.shape) == 2 or len(X.shape) == 3:

[jerryzh168]

If I'm not mistaken you only need to remove the 2 from

pytorch/test/quantization/core/test_quantized_op.py

Line 2106 in 6d45d7a

max_sides = (2, 3, 4, 5)

as well as the ==2 check here

pytorch/test/quantization/core/test_quantized_op.py

Line 2130 in 6d45d7a

if len(X.shape) == 2 or len(X.shape) == 3:

are you saying I don't need to add an extra test for batchnorm 1d? I can remove that
The goal for the test is to make sure batchnorm 1d outputs a Tensor with correct shape

[pmeier]

Oh sorry, I misunderstood the intent of the PR. Ignore my earlier comment. Before you land this PR though, could you locally put in assert qx.shape == qy.shape to make sure this is actually passing?

[jerryzh168]

Oh sorry, I misunderstood the intent of the PR. Ignore my earlier comment. Before you land this PR though, could you locally put in assert qx.shape == qy.shape to make sure this is actually passing?

yes, will do that

[jerryzh168]

@jerryzh168 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[pmeier]

LGTM! If it passes locally with the suggested check, it should be fine.

[jerryzh168]

LGTM! If it passes locally with the suggested check, it should be fine.

yes it did pass the shape check locally

[jerryzh168]

@jerryzh168 has imported this pull request. If you are a Facebook employee, you can view this diff on Phabricator.

[facebook-github-bot]

This pull request has been merged in 9de0c21.