[Torch][Quantized] Fix converting serialized quantized models #5839
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
siju-samuel
reviewed
Jun 18, 2020
siju-samuel
reviewed
Jun 18, 2020
siju-samuel
approved these changes
Jun 18, 2020
trevor-m
pushed a commit
to trevor-m/tvm
that referenced
this pull request
Jun 30, 2020
…#5839) * [Torch] Fix converting serialized quantized models * clean up dtype check * comment clean up
zhiics
pushed a commit
to neo-ai/tvm
that referenced
this pull request
Jul 2, 2020
…#5839) * [Torch] Fix converting serialized quantized models * clean up dtype check * comment clean up
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Outdated suggestions cannot be applied.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.
This is a workaround for the issue reported in pytorch/pytorch#39690
In short, if a quantized PyTorch model is serialized and loaded back, dtypes of output tensors are dropped and the loaded model doesn't have
QUInt8types at all.This becomes a problem when converting some Torch ops. For example, below the output dtype of
quantize_per_tensorbecomes float (Tensormeans float tensor, wrong), soaten::adaptive_avg_pool2dthinks this is a float operation. But obviously the output ofaten::quantize_per_tensorshould be a quantized tensor, soaten::adaptive_avg_pool2dhas to be converted to the quantized version.The quantized resnet in torchvision uses
aten::adaptive_avg_pool2d. So right now if we save and load back the qresnet, we get garbage result.please review @siju-samuel @anijain2305
cc @jjohnson-arm