[QNN][TFLite] TFLite rounding mode support #7456
Closed
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Outdated suggestions cannot be applied.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.
Almost work is done by @Fwd-IV Thanks! I just port it from #4828.
Some other import changes should be noticed is in the rewrite_simplify.cc / canonical_simplify.cc. We introduce expression complexity to determine whether we should do these optimizations. After introducing tflite rounding, we will have complex expressions, like this:
If we don't disable this , we even can not compile whole mobilenet v2 quantized model successfully.
This is also initial support for TFLite bit extract comparison . After this work, our qnn mobilenetv1/v2/inception and so on models we could get bit extract compared with tflite. I also think we have more work to do for more models.
Co-author-by: @Fwd-IV
@anijain2305 @mbrookhart @giuseros