app.py

import os
import shutil
import subprocess

# Clone and install dependencies
print("Cloning the repository...")
subprocess.run(["git", "clone", "https://huggingface.co/irotem98/edge_vlm"])
print("Installing dependencies...")
subprocess.run(["pip", "install", "-r", "edge_vlm/requirements.txt"])
subprocess.run(["pip", "install", "sentencepiece"])

# Copy all files from edge_vlm to current directory
print("Copying files...")
source_dir = "edge_vlm"
destination_dir = "."

import torch
import gradio as gr


for item in os.listdir(source_dir):
    source_item = os.path.join(source_dir, item)
    destination_item = os.path.join(destination_dir, item)

    if os.path.isdir(source_item):
        if os.path.exists(destination_item):
            shutil.rmtree(destination_item)
        shutil.copytree(source_item, destination_item)
    else:
        shutil.copy(source_item, destination_item)

print("Files copied successfully.")

# Now import the model from the copied files
from model import MoondreamModel

# Load the model and tokenizer
print("Loading model...")
model = MoondreamModel.load_model()
print("Model loaded.")
print("Loading tokenizer...")
tokenizer = MoondreamModel.load_tokenizer()
print("Tokenizer loaded.")

# Define the default question
default_question = "Describe the image."

# Function to handle image and return generated caption
def generate_caption_with_default(image):
    print("Preprocessing image...")
    preprocessed_image = MoondreamModel.preprocess_image(image)
    print("Image preprocessed.")
    
    print("Generating caption...")
    caption = MoondreamModel.generate_caption(model, preprocessed_image, tokenizer)
    print("Caption generated.")
    
    return caption

# Create Gradio interface
print("Setting up Gradio interface...")
interface = gr.Interface(
    fn=generate_caption_with_default,
    inputs=gr.Image(type="pil", label="Upload an Image"),
    outputs="text",
    title="Image Caption Generator",
    description=(
        f"The default question is: '{default_question}'.\n\n"
        "Please note that the inference may take up to 200 seconds due to long captions and CPU limitations.\n\n"
        "[![Hugging Face Model](https://img.shields.io/badge/Hugging%20Face-Model-blue)](https://huggingface.co/irotem98/edge_vlm) "
        "[![GitHub Repo](https://img.shields.io/badge/GitHub-Repo-green)](https://github.com/rotem154154/edge_vlm)"
    )
)
# Launch the interface
print("Launching interface...")
interface.launch()