Breaking News: Grepper is joining You.com. Read the official announcement!

pytorch load state dict for QAT model

JUNJIE ZHEN answered on August 29, 2022 Popularity 9/10 Helpfulness 1/10

answer pytorch load state dict for QAT model

pytorch load state dict for QAT model

Comment

import torch

...
self.create_net()
custom_load_state_dict(self.net, torch.load(str(filepath), map_location="cpu"))
self.net.qconfig = torch.quantization.get_default_qat_qconfig('fbgemm')
torch.quantization.prepare_qat(self.net, inplace=True)
# Freeze observers and bn immediately after model load
self.net.apply(torch.nn.intrinsic.qat.freeze_bn_stats)
self.net.apply(torch.quantization.disable_observer)
...

# Another file
def custom_load_state_dict(target_module: torch.nn.Module, source_state_dict: dict) -> None:
    target_state_dict = target_module.state_dict()
    for key, source_tensor in source_state_dict.items():
        if key in target_state_dict:
            if target_state_dict[key].shape == source_tensor.shape:
                target_state_dict[key] = source_tensor
    unmatched_keys = target_module.load_state_dict(target_state_dict)
    if unmatched_keys:
        print(f'Unmatched keys during model loading:\n{unmatched_keys}')

Popularity 9/10 Helpfulness 1/10 Language python

Source: stackoverflow.com

Tags: load model python pytorch state-dict

Link to this answer
Share Copy Link