azalea/CLAP
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[O] Allow backward-compatibility for torch state dict

Browse Source
This commit is contained in:
Azalea Gui
2023-10-11 18:54:51 -04:00
parent d47289931d
commit ea5629de26
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/CLAPWrapper.py
+1 -1
View File
@@ -98,7 +98,7 @@ class CLAPWrapper():
# We unwrap the DDP model and save. If the model is not unwrapped and saved, then the model needs to unwrapped before `load_state_dict`: # We unwrap the DDP model and save. If the model is not unwrapped and saved, then the model needs to unwrapped before `load_state_dict`:
# Reference link: https://discuss.pytorch.org/t/how-to-load-dataparallel-model-which-trained-using-multiple-gpus/146005 # Reference link: https://discuss.pytorch.org/t/how-to-load-dataparallel-model-which-trained-using-multiple-gpus/146005
clap.load_state_dict(model_state_dict) clap.load_state_dict(model_state_dict, strict=False)
clap.eval() # set clap in eval mode clap.eval() # set clap in eval mode
tokenizer = AutoTokenizer.from_pretrained(args.text_model) tokenizer = AutoTokenizer.from_pretrained(args.text_model)