Add op (hsatck & vstack) | feat(torchlib) #762

titaiwangms · 2023-06-01T23:20:22Z

Missing operator: https://github.com/microsoft/onnx-converters-private/issues/168
Follow up issue track: microsoft/onnxruntime#16296

onnxscript/function_libs/torch_lib/ops/core.py

titaiwangms · 2023-06-02T20:03:55Z

~~The op works in test_fx_op_consistency.py, so I figure there might be a bug inside graph_executor on Sequence inputs~~ The op is not even used in FX graph test in test_fx_op_consistency.py...

fatcat-z · 2023-06-05T10:20:45Z

Currently, for the issue that if/else block may return different shape/type, the alternative is to create 2 functions with different inputs for the same aten op, just like aten_bitwise_not. Could you please change this function as well?

Need microsoft/onnxscript#762 [ghstack-poisoned]

…767) op.SequenceMap comes to the rescue for element iteration cases. (Maybe it's just me that didn't know this...) This PR helps complete #762

onnxscript/function_libs/torch_lib/ops/core.py

Need microsoft/onnxscript#762 [ghstack-poisoned]

justinchuby · 2023-06-08T00:03:20Z

onnxscript/function_libs/torch_lib/ops/core.py

 @torch_op("aten::atleast_1d")
 def aten_atleast_1d(self: Sequence[TTensor]) -> TTensor:
    """atleast_1d(Tensor self) -> Tensor"""



Looks like it works.

Yes, it works. The error is something else. For this particular case, if we need to move aten::atleast_1d to a private share node, it would be a whole function...

I would then lift the whole function. Just don't call one aten in another because that will create dependency assumptions and limit our ability to refactor

justinchuby · 2023-06-08T18:29:20Z

onnxscript/function_libs/torch_lib/ops/core.py

    return op.SequenceMap(self, body=reshape_to_1d)


+@torch_op("aten::atleast_1d", trace_only=True)


Possible to make it not trace only, such that the function name is retained in the graph?

Would nested affect performance?

justinchuby · 2023-06-08T18:30:26Z

onnxscript/tests/function_libs/torch_lib/ops_test_data.py

    ),
+    xfail(
+        "vstack",
+        reason="fixme: A bug of constant-propagation optimization within the subgraph, we can avoid it by turning off graph-optimizations in session options",


Should we turn it off in test and file a bug to ORT?

~~Would nested affect performance?~~

xfail actually works. I think CI breaks because of something else. Actually @gramalingam spot the bug, but I can still file an issue to track.

microsoft/onnxruntime#16296

microsoft/onnxscript#762 is actually not used by FX graph. Fixes microsoft/onnx-converters-private#168 [ghstack-poisoned]

microsoft/onnxscript#762 is actually not used by FX graph. Fixes https://github.com/microsoft/onnx-converters-private/issues/168 Pull Request resolved: #102872 Approved by: https://github.com/justinchuby

Add op (hsatck) | feat (torchlib)

115f56d

titaiwangms added the module: torchlib Related to the torch/aten function lib in development label Jun 1, 2023

titaiwangms requested review from fatcat-z, justinchuby and xiaowuhu June 1, 2023 23:20

BowenBao reviewed Jun 1, 2023

View reviewed changes

onnxscript/function_libs/torch_lib/ops/core.py Outdated Show resolved Hide resolved

titaiwangms added 2 commits June 2, 2023 10:34

Merge branch 'main' into titaiwang/add_atleast_nd

1aa7e26

add logic to check first element

10a1c1a

titaiwangms added the help wanted Extra attention is needed label Jun 2, 2023

add vstack

5c311b5

titaiwangms changed the title ~~Add op (hsatck) | feat (torchlib)~~ Add op (hsatck & vstack) | feat (torchlib) Jun 2, 2023

remove print

7a70e32

titaiwangms mentioned this pull request Jun 2, 2023

[ONNX] Support aten::hstack and aten::vstack pytorch/pytorch#102872

Closed

justinchuby changed the title ~~Add op (hsatck & vstack) | feat (torchlib)~~ Add op (hsatck & vstack) | feat(torchlib) Jun 3, 2023

trace only the hstack

879ef75

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 6, 2023

Update base for Update on "[ONNX] Support aten::hstack and aten::vstack"

d9f277a

Need microsoft/onnxscript#762 [ghstack-poisoned]

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 6, 2023

Update on "[ONNX] Support aten::hstack and aten::vstack"

5d48fd1

Need microsoft/onnxscript#762 [ghstack-poisoned]

titaiwangms mentioned this pull request Jun 6, 2023

App op (atleast_1d and atleast_2d and atleast_3d) | feat (torchlib) #767

Merged

Merge branch 'main' into titaiwang/add_atleast_nd

fbe4ef5

titaiwangms commented Jun 6, 2023

View reviewed changes

onnxscript/function_libs/torch_lib/ops/core.py Outdated Show resolved Hide resolved

titaiwangms added 3 commits June 7, 2023 20:33

add atleast_Nd into hstack/vstack

06b3de2

Merge branch 'main' into titaiwang/add_atleast_nd

e17ce78

add atleast_Nd

0480fbd

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 7, 2023

Update base for Update on "[ONNX] Support aten::hstack and aten::vstack"

3cd4167

Need microsoft/onnxscript#762 [ghstack-poisoned]

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 7, 2023

Update on "[ONNX] Support aten::hstack and aten::vstack"

ba0c320

Need microsoft/onnxscript#762 [ghstack-poisoned]

justinchuby reviewed Jun 8, 2023

View reviewed changes

add xfail and shared func

2d25896

justinchuby reviewed Jun 8, 2023

View reviewed changes

titaiwangms added 2 commits June 8, 2023 11:30

Merge branch 'main' into titaiwang/add_atleast_nd

f3ba885

remove trace only

a24e620

justinchuby approved these changes Jun 8, 2023

View reviewed changes

titaiwangms merged commit 7f3da13 into microsoft:main Jun 8, 2023

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 13, 2023

Update on "[ONNX] Support aten::hstack and aten::vstack"

d1af8c9

microsoft/onnxscript#762 is actually not used by FX graph. Fixes microsoft/onnx-converters-private#168 [ghstack-poisoned]

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 13, 2023

Update on "[ONNX] Support aten::hstack and aten::vstack"

02eda0d

microsoft/onnxscript#762 is actually not used by FX graph. Fixes microsoft/onnx-converters-private#168 [ghstack-poisoned]

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 14, 2023

Update on "[ONNX] Support aten::hstack and aten::vstack"

cb350fd

microsoft/onnxscript#762 is actually not used by FX graph. Fixes microsoft/onnx-converters-private#168 [ghstack-poisoned]

titaiwangms added a commit to pytorch/pytorch that referenced this pull request Jun 15, 2023

Update on "[ONNX] Support aten::hstack and aten::vstack"

6e46b64

microsoft/onnxscript#762 is actually not used by FX graph. Fixes microsoft/onnx-converters-private#168 [ghstack-poisoned]

		return op.SequenceMap(self, body=reshape_to_1d)


		@torch_op("aten::atleast_1d", trace_only=True)

Add op (hsatck & vstack) | feat(torchlib) #762

Add op (hsatck & vstack) | feat(torchlib) #762

Uh oh!

Conversation

titaiwangms commented Jun 1, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

titaiwangms commented Jun 2, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

fatcat-z commented Jun 5, 2023

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

justinchuby Jun 8, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

titaiwangms Jun 8, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

titaiwangms commented Jun 1, 2023 •

edited

Loading

titaiwangms commented Jun 2, 2023 •

edited

Loading

justinchuby Jun 8, 2023 •

edited

Loading

titaiwangms Jun 8, 2023 •

edited

Loading