feat: introduce type propagation infrastructure #2469
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This PR introduces type propagation infrastructure through pytorch inference. Inferred output types are used to set the TensorRT engines output types.
github-actions
bot
added
component: conversion
7 tasks
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?s=60&v=4){kind=small}
| @@ -340,6 +340,19 @@ void getSegmentsOutputByRunning( | |||
|
|
|||
| seg_block.register_inshapes(input_shapes, shape_mode); | |||
| seg_block.register_intypes(input_types); | |||
|
|
|||
| // get output type for each segmented block so this can be used in conversion process | |||
| std::vector<at::ScalarType> output_types; | |||
There was a problem hiding this comment.
One thing to check: Because we're running this at the end of shape propagation does this work for fully convertible modules?
From memory I think that shape prop is only called from BuildHybridGraph->partition() which I don't think is called if all ops are convertible.
| @@ -265,6 +269,7 @@ void MarkOutputs(ConversionCtx* ctx, at::ArrayRef<const torch::jit::Value*> outp | |||
| } | |||
|
|
|||
| if (!setOutput) { | |||
| out_tensor->setType(util::ScalarTypeToTRTDataType(out_types[out_idx++])); | |||
There was a problem hiding this comment.
The out_idx needs to be incremented even if one of the outputs is an input (see where setOutput may be set above)
gs-olive
reviewed
Dec 15, 2023
There was a problem hiding this comment.
Overall, looks good, just needs a few small fixes. Needs a test case for verification. Additionally, it seems some of the Torchscript test cases are failing due to this change:
ERROR conda.cli.main_run:execute(41): `conda run python -m pytest --junitxml=/tmp/test_results/ts_api_test_results.xml api/` failed. (See above for error)
Fatal Python error: Segmentation fault
Current thread 0x00007f4711145740 (most recent call first):
File "/opt/python/cp38-cp38/lib/python3.8/site-packages/torch_tensorrt/ts/_compiler.py", line 266 in convert_method_to_trt_engine
File "/__w/TensorRT/TensorRT/pytorch/tensorrt/tests/py/ts/api/test_classes.py", line 300 in test_detect_invalid_input_binding
This may be due to the reason in the comment above #2469 (comment)
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Description
This PR introduces type propagation infrastructure through pytorch inference.
Inferred output types are used to set the TensorRT engines output types.
Related PR: #1853
Previous solution has several limitations.
Checklist: