README.md

---
license: cc-by-nc-sa-4.0
language:
- en
pipeline_tag: image-feature-extraction
tags:
- pathology
- neuropathology
- foundation_model
- vit
---

# neuroFM_HE20x

ViT-large (300M parameters) trained on a diverse neuropathology dataset.


## Model Usage

To get started, first clone the repository with this command:
```bash
  git clone --no-checkout https://huggingface.co/MountSinaiCompPath/neuroFM_HE20x && cd neuroFM_HE20x && git sparse-checkout init --no-cone && git sparse-checkout set '/*' '!*.bin' && git checkout
```

Now you can use the following code:
```python
from PIL import Image
import numpy as np
import vision_transformer
import torch
import torch.nn as nn
import torchvision.transforms as transforms
from huggingface_hub import PyTorchModelHubMixin

class neuroFM_HE20x(nn.Module, PyTorchModelHubMixin):
    def __init__(self):
        super().__init__()
	vit_kwargs = dict(
            img_size=224,
            patch_size=14,
            init_values=1.0e-05,
            ffn_layer='swiglufused',
            block_chunks=4,
            qkv_bias=True,
            proj_bias=True,
            ffn_bias=True,
        )
        self.encoder = vision_transformer.__dict__['vit_large'](**vit_kwargs)
    
    def forward(self, x):
        return self.encoder(x)

# Download model
model = neuroFM_HE20x.from_pretrained("MountSinaiCompPath/neuroFM_HE20x")

# Set up transform
transform = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
])

# Image
img = np.random.randint(0, 256, size=224*224*3).reshape(224,224,3).astype(np.uint8)
img = Image.fromarray(img)
img = transform(img).unsqueeze(0)

# Inference
with torch.no_grad():
    h = model(img)
```