search

facebookresearch / d2go

D2Go is a toolkit for efficient deep learning
Apache License 2.0
826 stars 197 forks source link

allow to ignore state dict keys in QAT model #642

Closed stephenyan1231 closed 6 months ago

stephenyan1231 commented 6 months ago

Summary: When we build a QAT model using FX graph mode API prepare_qat_fx and convert_fx, they will run symbolic tracing following module.forward().

In certain cases, such as a module takes constant tensor input, the symbolic tracing will add new tensor attributes with name prefix _tensor_constant (https://fburl.com/code/msc4ch4o), which becomes new keys in the QAT model state dict.

In current implementation of _setup_non_qat_to_qat_state_dict_map, it asserts # of keys in the state dict of original- and QAT model should be the same.

Thus, we extend qat_state_dict_keys_to_ignore method by adding an argument, which allows to ignore specified state dict keys in the QAT model.

Differential Revision: D52152706

facebook-github-bot commented 6 months ago

This pull request was exported from Phabricator. Differential Revision: D52152706

facebook-github-bot commented 6 months ago

This pull request was exported from Phabricator. Differential Revision: D52152706

facebook-github-bot commented 6 months ago

This pull request was exported from Phabricator. Differential Revision: D52152706

facebook-github-bot commented 6 months ago

This pull request was exported from Phabricator. Differential Revision: D52152706

facebook-github-bot commented 6 months ago

This pull request has been merged in facebookresearch/d2go@c2256758202eb51ae8f21200f58dcbb70ca96690.