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: https://github.com/evolutionaryscale/esm/blob/39a3a6cb1e722347947dc375e3f8e2ba80ed8b59/esm/utils/constants/esm3.py#L100-L105

[zhang0730]

Did anybody success? I tried:

        model_path = "/disk2/044/scprint/temp/data/esmc-600m-2024-12"
        
        # Check if the model path exists
        if not os.path.exists(model_path):
            raise FileNotFoundError(f"Model path not found: {model_path}")
        
        # Load the model configuration
        config_path = os.path.join(model_path, "config.json")
        config = AutoConfig.from_pretrained(config_path)
        
        # Load the tokenizer
        tokenizer = AutoTokenizer.from_pretrained(model_path)
        
        # Initialize the model
        client = ESMC(
            n_heads=config.num_attention_heads,
            n_layers=config.num_hidden_layers,
            tokenizer=tokenizer
        ).to("cuda" if cuda else "cpu")
        
        # Load the model weights
        weights_path = os.path.join(model_path, "/data/weights/esmc_600m_2024_12_v0.pth")

but it noted that OSError: Can't load tokenizer for '/disk2/044/scprint/temp/data/esmc-600m-2024-12'. I screened the https://huggingface.co/EvolutionaryScale/esmc-600m-2024-12/tree/main, but not found tokenizer.json、vocab.json etc.

[zhang0730]

I solved the problem by both modifying the data_root() and model = ESM3.from_pretrained() like this:

---

def data_root(model: str):
    if "INFRA_PROVIDER" in os.environ:
        return Path("")
    # Try to download from hugginface if it doesn't exist
    if model.startswith("esm3"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esm3-sm-open-v1"))
    elif model.startswith("esmc-300"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esmc-300m-2024-12"))
    elif model.startswith("esmc-600"):
        path = Path("/disk2/044/scprint/temp/data/esmc-600m-2024-12") ###manually

---

    local_weights_path = "/disk2/044/scprint/temp/data/esmc-600m-2024-12/data/weights/esmc_600m_2024_12_v0.pth"
    client = load_local_model(
        "esmc_600m",
        device="cuda" if cuda else "cpu"
     )
     state_dict = torch.load(local_weights_path, map_location="cuda" if cuda else "cpu")
     client.load_state_dict(state_dict)

---

Hope this helps.

[kuku609]

I solved the problem by both modifying the data_root() and model = ESM3.from_pretrained() like this:

def data_root(model: str):
    if "INFRA_PROVIDER" in os.environ:
        return Path("")
    # Try to download from hugginface if it doesn't exist
    if model.startswith("esm3"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esm3-sm-open-v1"))
    elif model.startswith("esmc-300"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esmc-300m-2024-12"))
    elif model.startswith("esmc-600"):
        path = Path("/disk2/044/scprint/temp/data/esmc-600m-2024-12") ###manually

    local_weights_path = "/disk2/044/scprint/temp/data/esmc-600m-2024-12/data/weights/esmc_600m_2024_12_v0.pth"
    client = load_local_model(
        "esmc_600m",
        device="cuda" if cuda else "cpu"
     )
     state_dict = torch.load(local_weights_path, map_location="cuda" if cuda else "cpu")
     client.load_state_dict(state_dict)

Hope this helps.

requests.exceptions.SSLError: (MaxRetryError('HTTPSConnectionPool(host='huggingface.co', port=443): Max retries exceeded with url: /api/models/EvolutionaryScale/esmc-600m-2024-12/revision/main (Caused by SSLError(SSLCertVerificationError(1, "[SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: Hostname mismatch, certificate is not valid for 'huggingface.co'. (_ssl.c:1007)")))'), '(Request ID: 119e2e20-ed26-4b45-a398-2e1266a498d0)')

您好，感谢您分享的解决办法，请问您有遇到这个问题吗？

[xinalbert]

firstly, clone all the weights from huggingface to your local machine. such as: git clone https://user:[email protected]/EvolutionaryScale/esmc-300m-2024-12 all of the three categories should be downloaded.

then, change the path in the function data_root() to your local path.

# original_code
def data_root(model: str):
    if "INFRA_PROVIDER" in os.environ:
        return Path("")
    # Try to download from hugginface if it doesn't exist
    if model.startswith("esm3"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esm3-sm-open-v1"))
    elif model.startswith("esmc-300"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esmc-300m-2024-12"))
    elif model.startswith("esmc-600"):
        path = Path(snapshot_download(repo_id="EvolutionaryScale/esmc-600m-2024-12"))
    else:
        raise ValueError(f"{model=} is an invalid model name.")
    return path

# change to
def data_root(model: str):
    if "INFRA_PROVIDER" in os.environ:
        return Path("")
    # Try to download from hugginface if it doesn't exist
    if model.startswith("esm3"):
        path = Path('/your/path/esm3-sm-open-v1')
    elif model.startswith("esmc-300"):
        path = Path('/your/path/esmc-300m-2024-12')
    elif model.startswith("esmc-600"):
        path = Path("/your/path/esmc-600m-2024-12")
    else:
        raise ValueError(f"{model=} is an invalid model name.")
    return path

and then test the code!

# from huggingface_hub import login
from esm.models.esm3 import ESM3
from esm.sdk.api import ESM3InferenceClient, ESMProtein, GenerationConfig

model: ESM3InferenceClient = ESM3.from_pretrained("esm3-open").to("cuda") # or 
prompt = '''MDEQSQGMQGPPVTQFQPQKALRPDMGYNTLANFRIEKKIGRGQFSEVYRASCLLDGVPVALKKVQIFDLMDAKARADCIKEIDLLKQLNHPNVIKYYASFIEDNELNIVLELADAGDLSRMIKHFKKQKRLIPERTVWKYFVQLCSALDHMHSRRVMHRDIKPANVFITATGVVKLGDLGLGRF'''
protein = ESMProtein(sequence=prompt)
# Generate the sequence, then the structure. This will iteratively unmask the sequence track.
protein = model.generate(protein, GenerationConfig(track="sequence", num_steps=8, temperature=0.7))

# We can show the predicted structure for the generated sequence.
protein = model.generate(protein, GenerationConfig(track="structure", num_steps=8))
protein.to_pdb("test.pdb")
# # Then we can do a round trip design by inverse folding the sequence and recomputing the structure
# protein.sequence = None
# protein = model.generate(protein, GenerationConfig(track="sequence", num_steps=8))
# protein.coordinates = None
# protein = model.generate(protein, GenerationConfig(track="structure", num_steps=8))
# protein.to_pdb("./round_tripped.pdb")