from_pretrained() Error. Cannot use a path to import local model?

[1830416002]

my code like this:

model = ESM3.from_pretrained(model_name = "./huggingface/hub/models--EvolutionaryScale--esm3-sm-open-v1/snapshots/66ecd636588d3100e13598a5720678db6583d01c/", device=torch.device("cpu"))

and give the error:
`---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In [27], line 1
----> 1 model = ESM3.from_pretrained(model_name = "./huggingface/hub/models--EvolutionaryScale--esm3-sm-open-v1/snapshots/66ecd636588d3100e13598a5720678db6583d01c/", device=torch.device("cpu"))

File ~/users/wtt/anaconda3/envs/esm3/lib/python3.11/site-packages/esm/models/esm3.py:251, in ESM3.from_pretrained(cls, model_name, device)
249 if device is None:
250 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
--> 251 model = load_local_model(model_name, device=device)
252 if device.type != "cpu":
253 model = model.to(torch.bfloat16)

File ~/users/wtt/anaconda3/envs/esm3/lib/python3.11/site-packages/esm/pretrained.py:87, in load_local_model(model_name, device)
83 def load_local_model(
84 model_name: str, device: torch.device = torch.device("cpu")
85 ) -> nn.Module:
86 if model_name not in LOCAL_MODEL_REGISTRY:
---> 87 raise ValueError(f"Model {model_name} not found in local model registry.")
88 return LOCAL_MODEL_REGISTRYmodel_name

ValueError: Model ./huggingface/hub/models--EvolutionaryScale--esm3-sm-open-v1/snapshots/66ecd636588d3100e13598a5720678db6583d01c/ not found in local model registry.`

[Jaychonuo]

you may further refer .../66ecd636588d3100e13598a5720678db6583d01c/data

[1830416002]

I have tried and have same error:
`---------------------------------------------------------------------------
ValueError Traceback (most recent call last)
Cell In [2], line 1
----> 1 model = ESM3.from_pretrained(model_name = "./huggingface/hub/models--EvolutionaryScale--esm3-sm-open-v1/snapshots/66ecd636588d3100e13598a5720678db6583d01c/data", device=torch.device("cpu"))

File ~/users/wtt/anaconda3/envs/esm3/lib/python3.11/site-packages/esm/models/esm3.py:251, in ESM3.from_pretrained(cls, model_name, device)
249 if device is None:
250 device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
--> 251 model = load_local_model(model_name, device=device)
252 if device.type != "cpu":
253 model = model.to(torch.bfloat16)

File ~/users/wtt/anaconda3/envs/esm3/lib/python3.11/site-packages/esm/pretrained.py:87, in load_local_model(model_name, device)
83 def load_local_model(
84 model_name: str, device: torch.device = torch.device("cpu")
85 ) -> nn.Module:
86 if model_name not in LOCAL_MODEL_REGISTRY:
---> 87 raise ValueError(f"Model {model_name} not found in local model registry.")
88 return LOCAL_MODEL_REGISTRYmodel_name

ValueError: Model ./huggingface/hub/models--EvolutionaryScale--esm3-sm-open-v1/snapshots/66ecd636588d3100e13598a5720678db6583d01c/data not found in local model registry.`

[LHucass]

The same question for me
My remote server could not get connection with Internet so I must load the local model.

[Jaychonuo]

The same question for me My remote server could not get connection with Internet so I must load the local model.

https://blog.csdn.net/m0_61474277/article/details/140348032
for reference

[ebetica]

This is the set of model names that the function supports: https://github.com/evolutionaryscale/esm/blob/main/esm/utils/constants/models.py#L2-L8

If you're looking to download it without internet, I think you should refer to huggingface for details. I would try to just put the right files in the right locations, or to modify this file:

esm/esm/utils/constants/esm3.py

Lines 100 to 105 in 39a3a6c

	def data_root():
	if "INFRA_PROVIDER" in os.environ:
	return Path("")
	# Try to download from hugginface if it doesn't exist
	path = Path(snapshot_download(repo_id="EvolutionaryScale/esm3-sm-open-v1"))
	return path