app.py

import os
import torch
from transformers import AutoImageProcessor, ViTForImageClassification, pipeline
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
import gradio as gr

MODEL_REPO = "YUGOROU/Soba-or-Udon-vit"
MODEL_FILE = "model.safetensors"
BASE_MODEL = "google/vit-base-patch16-224"


def load_model():
    processor = AutoImageProcessor.from_pretrained(BASE_MODEL)
    model = ViTForImageClassification.from_pretrained(
        BASE_MODEL,
        num_labels=3,
        ignore_mismatched_sizes=True
    )
    weights_path = hf_hub_download(repo_id=MODEL_REPO, filename=MODEL_FILE)
    state_dict = load_file(weights_path)
    model_state = model.state_dict()
    filtered_state = {
        k: v for k, v in state_dict.items()
        if k in model_state and v.shape == model_state[k].shape
    }
    model.load_state_dict(filtered_state, strict=False)
    model.eval()
    device = 0 if torch.cuda.is_available() else -1
    model.config.id2label = {
        0: "うどん",
        1: "蕎麦",
        2: "どちらでもない",
    }
    model.config.label2id = {v: k for k, v in model.config.id2label.items()}
    pipe = pipeline(
        "image-classification",
        model=model,
        image_processor=processor,
        device=device,
    )
    return pipe


pipe = load_model()

def predict(image):
    result = pipe(image)[0]
    return {result["label"]: result["score"]}

demo = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs=gr.Label(num_top_classes=2),
    title="Soba or Udon Classifier",
    description="Upload an image of noodles to find out whether it is Soba or Udon.",
)

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