None: Replace the Tensor Layer name from "Unnamed Layer #" to its original name #2329

wu6u3tw · 2023-09-20T20:11:20Z

Description

To improve the debuggability, we try to set the layer name for each myelin output for a more meaningful naming so that we can replace the "Unnamed Layer #" with its original name.

Fixes # (issue)

Type of change

Please delete options that are not relevant and/or add your own.

Bug fix (non-breaking change which fixes an issue)
New feature (non-breaking change which adds functionality)
Breaking change (fix or feature that would cause existing functionality to not work as expected)
This change requires a documentation update

Checklist:

My code follows the style guidelines of this project (You can use the linters)
I have performed a self-review of my own code
I have commented my code, particularly in hard-to-understand areas and hacks
I have made corresponding changes to the documentation
I have added tests to verify my fix or my feature
New and existing unit tests pass locally with my changes
I have added the relevant labels to my PR in so that relevant reviewers are notified

wu6u3tw · 2023-09-20T20:11:58Z

@nvpohanh for viz

gs-olive

This change would be very beneficial for layer naming, thanks for the PR! It would be preferable to add this change to the Dynamo directory instead. Could you copy the function set_layer_name with your changes to the file py/torch_tensorrt/dynamo/conversion/converter_utils.py, and then update imports of set_layer_name in any file within the py/torch_tensorrt/dynamo/ directory to use this version of the function? The fx/ directory should preferably be unchanged.

gs-olive · 2023-09-21T00:51:29Z

py/torch_tensorrt/fx/converters/converter_utils.py

+
+    for i in range(layer.num_outputs):
+        output = layer.get_output(i)
+        layer.name.append(f"-[{output.name}]")


Should layer.name.append(f"-[{output.name}]") instead be layer.name += f"-[{output.name}]"?

@wu6u3tw I think we should set the output tensor's name. Something like: output.name = "..."

We should not just append all the output tensor names to layer name because that makes layer name very long

nvpohanh · 2023-09-21T01:31:03Z

py/torch_tensorrt/fx/converters/converter_utils.py

        else f"{source_ir}_ops.{target.__name__}"
    )
-    layer.name = f"[{layer.type.name}]-[{target_name}]-[{name}]"
+    layer.name = f"[{layer.type.name}]-[{target_name}]-[{name}]-[#_of_outputs_{layer.num_outputs}]"


Please also set the layer metadata: layer.metadata = ...

Consider moving [#_of_outputs_{layer.num_outputs}] into metadata instead of in layer name directly

Co-authored-by: gs-olive <113141689+gs-olive@users.noreply.github.com>

…er (pytorch#2321)

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

Dynamo converter support for torch.ops.aten.erf.default op

pytorch#2339)

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

Signed-off-by: Bo Wang <bowa@nvidia.com>

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

wu6u3tw · 2023-10-09T20:19:33Z

@gs-olive thanks! I update the PR. Please let me know if there is any other advices.

…et_name

set layer name

1a599fa

facebook-github-bot added cla signed fx labels Sep 20, 2023

github-actions bot added component: api [Python] Issues re: Python API component: fx labels Sep 20, 2023

github-actions bot requested a review from gs-olive September 20, 2023 20:11

gs-olive suggested changes Sep 21, 2023

View reviewed changes

nvpohanh reviewed Sep 21, 2023

View reviewed changes

apbose and others added 21 commits September 21, 2023 12:03

FX converter documentation (pytorch#2039)

c875c39

aten::split converter (pytorch#2232)

19aabdd

Co-authored-by: gs-olive <113141689+gs-olive@users.noreply.github.com>

DLFW changes (pytorch#2281)

0a939df

feat: Add ATen lowering pass system (pytorch#2280)

ff4d940

fix: Support non -1 end idx and <0 start idx in aten::flatten convert…

65feab1

…er (pytorch#2321)

docs: [Automated] Regenerating documenation for 65feab1

e6e8099

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

support for torch.ops.aten.erf.default op

3c4c2fe

feat: support Dynamo converter for torch.ops.aten.erf.default op

670d2be

Dynamo converter support for torch.ops.aten.erf.default op

fix: Update Torchvision version to address dependency resolution issue (

ecdc040

pytorch#2339)

fix: Remove input aliasing of builtin ops (pytorch#2276)

7daa112

docs: [Automated] Regenerating documenation for 7daa112

b2aa255

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

fix: Allow low rank inputs in Python Runtime (pytorch#2282)

1033dff

docs: [Automated] Regenerating documenation for 1033dff

76de80d

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

fix: Address multi-GPU issue in engine deserialize (pytorch#2325)

338e542

docs: [Automated] Regenerating documenation for 338e542

117161a

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

feat: support deconv (1d, 2d, and Nd) dynamo converter (pytorch#2337)

251405d

docs: [Automated] Regenerating documenation for 251405d

a2a983b

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

Update usage of PyTorch's custom op API (pytorch#2193)

bece720

docs: [Automated] Regenerating documenation for bece720

78f2721

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

feat: support bmm converter in dynamo (pytorch#2248)

765933a

Signed-off-by: Bo Wang <bowa@nvidia.com>

docs: [Automated] Regenerating documenation for 765933a

0d402fb

Signed-off-by: Torch-TensorRT Github Bot <torch-tensorrt.github.bot@nvidia.com>

gs-olive and others added 3 commits October 6, 2023 16:18

minor fix: Update get_ir prefixes (pytorch#2369)

18dcdd0

Dynamo converter cat (pytorch#2343)

83176fe

update fix

4ceb796

github-actions bot added component: conversion Issues re: Conversion stage component: dynamo Issues relating to the `torch.compile` or `torch._dynamo.export` paths labels Oct 7, 2023

format

3d1be7e

wu6u3tw added 5 commits October 9, 2023 21:02

set layer name

2153cb9

update naming

3a0513a

update fix

c499d94

format

910e014

Merge branch 'dev-set_name' of github.com:wu6u3tw/TensorRT into dev-s…

a37ca1e

…et_name

wu6u3tw added 8 commits October 9, 2023 21:26

set layer name

f395f83

update naming

4b6b5f4

update fix

e085f20

format

917e6ab

set layer name

2359c0b

update naming

db547b5

update fix

88e1c2c

fix

d4d1c41

wu6u3tw closed this Oct 9, 2023

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

None: Replace the Tensor Layer name from "Unnamed Layer #" to its original name #2329

None: Replace the Tensor Layer name from "Unnamed Layer #" to its original name #2329

Uh oh!

wu6u3tw commented Sep 20, 2023

Uh oh!

wu6u3tw commented Sep 20, 2023

Uh oh!

gs-olive left a comment

Uh oh!

gs-olive Sep 21, 2023

Uh oh!

nvpohanh Sep 21, 2023

Uh oh!

nvpohanh Sep 21, 2023

Uh oh!

nvpohanh Sep 21, 2023

Uh oh!

gs-olive Sep 22, 2023

Uh oh!

wu6u3tw commented Oct 9, 2023 •

edited

Loading

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

None: Replace the Tensor Layer name from "Unnamed Layer #" to its original name #2329

None: Replace the Tensor Layer name from "Unnamed Layer #" to its original name #2329

Uh oh!

Conversation

wu6u3tw commented Sep 20, 2023

Description

Type of change

Checklist:

Uh oh!

wu6u3tw commented Sep 20, 2023

Uh oh!

gs-olive left a comment

Choose a reason for hiding this comment

Uh oh!

gs-olive Sep 21, 2023

Choose a reason for hiding this comment

Uh oh!

nvpohanh Sep 21, 2023

Choose a reason for hiding this comment

Uh oh!

nvpohanh Sep 21, 2023

Choose a reason for hiding this comment

Uh oh!

nvpohanh Sep 21, 2023

Choose a reason for hiding this comment

Uh oh!

gs-olive Sep 22, 2023

Choose a reason for hiding this comment

Uh oh!

wu6u3tw commented Oct 9, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

10 participants

wu6u3tw commented Oct 9, 2023 •

edited

Loading