import gradio as gr
from huggingface_hub import hf_hub_download
import torch, open_clip
from PIL import Image
from IPython.display import display

for model_name in ['RN50', 'ViT-B-32', 'ViT-L-14']:
    checkpoint_path = hf_hub_download("chendelong/RemoteCLIP", f"RemoteCLIP-{model_name}.pt", cache_dir='checkpoints')
    print(f'{model_name} is downloaded to {checkpoint_path}.')

model_name = 'RN50' # 'RN50' or 'ViT-B-32' or 'ViT-L-14'
model, _, preprocess = open_clip.create_model_and_transforms(model_name)
tokenizer = open_clip.get_tokenizer(model_name)

path_to_your_checkpoints = 'checkpoints/models--chendelong--RemoteCLIP/snapshots/bf1d8a3ccf2ddbf7c875705e46373bfe542bce38'

ckpt = torch.load(f"{path_to_your_checkpoints}/RemoteCLIP-{model_name}.pt", map_location="cpu")

def remote_clip(input_image,input_text):

    text_queries = [input_text]
    text = tokenizer(text_queries)
    
    image = Image.open(input_image)
    image = preprocess(image).unsqueeze(0)

    with torch.no_grad(), torch.cuda.amp.autocast():
        image_features = model.encode_image(image.cuda())
        text_features = model.encode_text(text.cuda())
        image_features /= image_features.norm(dim=-1, keepdim=True)
        text_features /= text_features.norm(dim=-1, keepdim=True)

        text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1).cpu().numpy()[0]

    print(f'Predictions of {model_name}:')
    for query, prob in zip(text_queries, text_probs):
        print(f"{query:<40} {prob * 100:5.1f}%")


demo = gr.Interface(fn=greet, inputs=[gr.Image(type="pil"), gr.Text(type="text")], outputs="text")
    
demo.launch()