[AFQ] Optimize tensor_flatten for runtime #1114
Open
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Outdated suggestions cannot be applied.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.
Stack from ghstack (oldest at bottom):
tensor_flatten is called at runtime when inputs are subclasses e.g. AffineQuantizedTensor
In case of using
float8_dynamic_activation_float8_weight
quantization, the activations will also be AQT.There was external complain, that unwrap_tensor_subclasses sometimes exceeds in duration that compiled region execution.
Profiling shows that tensor_flatten attribute access will go through torch_function handling (AQT has it).
If to remove this torch_function dispatch for each getattr at runtime - tensor_flatten becomes (in my measurements x9 faster)