wangyi111

/

Copernicus-FM

like 3

It doesn't seem like the model is compatible with the Transformers library.

Btw, you could consider making your CopernicusFMViT class inherit from the PyTorchModelHubMixin class. This way, you can do:

from src.model_vit import CopernicusFMViT
from huggingface_hub import hf_hub_download
import torch

# load model
model = CopernicusFMViT(...)

# equip with weights
filepath = hf_hub_download(repo_id="wangyi111/Copernicus-FM", filename="CopernicusFM_ViT_base_varlang_e100.pth")
state_dict = torch.load(filepath, map_location="cpu")
model.load_state_dict(state_dict)

# push to the hub
model.push_to_hub("wangyi111/Copernicus-FM-base")

# now anyone can use it like so
model = CopernicusFMViT.from_pretrained("wangyi111/Copernicus-FM-base")

This comes with a few benefits:

• download metrics
• safetensors for weights serialization.

We recommend pushing the base and large variants to separate model repositories (this way you will get download metrics for each).