[FX][AWQ][Scale Estimation][Mixed Precision] Add Data Aware Algorithm Support for FX Backend #3409
Conversation
github-actions
bot
added
NNCF PT
|
@anzr299 , please share conformance test results for these methods |
| for node in next_nodes: | ||
| target_points.append( | ||
| PTTargetPoint( | ||
| PTWeightCompressionAlgoBackend.TARGET_TYPE_TO_PT_INS_TYPE_MAP[TargetType.PRE_LAYER_OPERATION], |
There was a problem hiding this comment.
Could you please specify the actual target type without the dict?
There was a problem hiding this comment.
Done, my reasoning behind this approach was that it will be less modular if we directly define it in case of changes to the target type in the future
| class FXMixedPrecisionAlgoBackend(MixedPrecisionAlgoBackend, FXWeightCompressionAlgoBackend): | ||
| @staticmethod |
There was a problem hiding this comment.
Do we have a plan on how to reuse the same classes across all the PyTorch backends? If not, could you please create a ticket for that?
There was a problem hiding this comment.
Maybe we can set a flag in a common backend class (such as FX=False) when initializing it.
There was a problem hiding this comment.
@anzr299, It would be great if you would open a ticket and implement a make a proposal which we could discuss with the team
There was a problem hiding this comment.
Alright, I will do that
Changes
The following algorithms were added for Torch FX backend:
Tests
TemplateWeightCompressionwas extended for torch FX backend for the above algorithms intests/torch/fx/test_compress_weights.py