load_pretrained_model(model=model, model_path="", modules_to_load=None, exclude_modules="")
(model, model_path, modules_to_load=None, exclude_modules=None)
| 35 | |
| 36 | # Load and save |
| 37 | def load_pretrained_model(model, model_path, modules_to_load=None, exclude_modules=None): |
| 38 | ''' |
| 39 | load_pretrained_model(model=model, model_path="", |
| 40 | modules_to_load=None, exclude_modules="") |
| 41 | ''' |
| 42 | model_dict = torch.load(model_path, map_location=lambda storage, loc: storage) |
| 43 | if exclude_modules: |
| 44 | for e in exclude_modules.split(","): |
| 45 | model_dict = {k: v for k, v in model_dict.items() if not k.startswith(e)} |
| 46 | |
| 47 | if not modules_to_load: |
| 48 | src_dict = model_dict |
| 49 | else: |
| 50 | src_dict = {} |
| 51 | for module in modules_to_load.split(","): |
| 52 | src_dict.update({k: v for k, v in model_dict.items() if k.startswith(module)}) |
| 53 | |
| 54 | dst_state = model.state_dict() |
| 55 | dst_state.update(src_dict) |
| 56 | model.load_state_dict(dst_state) |
| 57 | def torch_save(model, save_path, optimizer=None, local_rank=0): |
| 58 | if local_rank != 0: |
| 59 | return |
no test coverage detected