import torch, torch.nn as nn
import torchvision.transforms as T
from PIL import Image
from huggingface_hub import hf_hub_download
import gradio as gr

# Download model
ckpt_path = hf_hub_download(repo_id="bit-wander/mnist-demo", filename="model.pt")

class SimpleCNN(nn.Module):
    def __init__(self):
        super().__init__()
        self.net = nn.Sequential(
            nn.Conv2d(1,16,3,padding=1), nn.ReLU(),
            nn.MaxPool2d(2),
            nn.Conv2d(16,32,3,padding=1), nn.ReLU(),
            nn.AdaptiveAvgPool2d(1),
            nn.Flatten(),
            nn.Linear(32,10)
        )
    def forward(self,x): return self.net(x)

model = SimpleCNN()
ckpt = torch.load(ckpt_path, map_location="cpu")
model.load_state_dict(ckpt['model_state_dict'])
model.eval()

transform = T.Compose([
    T.Grayscale(num_output_channels=1),
    T.Resize((28,28)),
    T.ToTensor(),
    T.Normalize((0.5,), (0.5,))
])

def predict(img: Image.Image):
    x = transform(img).unsqueeze(0)
    with torch.no_grad():
        logits = model(x)
        probs = torch.softmax(logits, dim=-1).squeeze().tolist()
    return {str(i): probs[i] for i in range(10)}

demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=3),
    title="MNIST Demo"
)

if __name__ == "__main__":
    demo.launch()