DeepWok / mase

Machine-Learning Accelerator System Exploration Tools
Other
105 stars 52 forks source link

Error when saving parametrized module #60

Open Linlangxie opened 3 months ago

Linlangxie commented 3 months ago

When saving a parametrized module, for my case, adding a torch.nn.utils.parametrize.register_parametrization() to the layers, it raises error: RuntimeError: Serialization of parametrized modules is only supported through state_dict(), at /mase/machop/chop/passes/graph/interface/save_and_load.py line 13, save_graph_module_ckpt(). It seems like torch.save() cannot support the parametrized module to save the model structure. 截图 2024-03-13 15-56-27