app.py

import spaces
import gradio as gr
from transformers import AutoModelForCausalLM, AutoProcessor, GenerationConfig
from PIL import Image
import torch
import requests

# Load the processor and model
processor = AutoProcessor.from_pretrained(
    'allenai/Molmo-7B-D-0924',
    trust_remote_code=True,
    torch_dtype='auto',
    device_map='auto'
)

model = AutoModelForCausalLM.from_pretrained(
    'allenai/Molmo-7B-D-0924',
    trust_remote_code=True,
    torch_dtype='auto',
    device_map='auto'
)

@spaces.GPU
def describe_image(image, prompt):
    # Process the image with the user-provided text prompt
    inputs = processor.process(images=[image], text=prompt)
    
    # Move inputs to the correct device and make a batch of size 1
    inputs = {k: v.to(model.device).unsqueeze(0) for k, v in inputs.items()}
    
    # Generate output with a maximum of 200 new tokens
    output = model.generate_from_batch(
        inputs,
        GenerationConfig(max_new_tokens=200, stop_strings="<|endoftext|>"),
        tokenizer=processor.tokenizer
    )
    
    # Decode and return the generated text
    generated_tokens = output[0, inputs['input_ids'].size(1):]
    generated_text = processor.tokenizer.decode(generated_tokens, skip_special_tokens=True)
    
    return generated_text

# Gradio interface using the latest API
with gr.Blocks() as demo:
    gr.Markdown("# Visual Language Model - Molmo")
    with gr.Row():
        image_input = gr.Image(type="pil", label="Upload an image")
        text_input = gr.Textbox(label="Enter a prompt", placeholder="Describe this image...")
    output_text = gr.Textbox(label="Generated Description")
    submit_button = gr.Button("Generate Description")
    
    # Connect the inputs (image, text prompt) to the function and output
    submit_button.click(fn=describe_image, inputs=[image_input, text_input], outputs=output_text)

# Launch the app
demo.launch()