app.py

import spaces
import gradio as gr
import torch
from transformers import Qwen2VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info

device = "cuda" if torch.cuda.is_available() else "cpu"

MODEL_REPO = "Qwen/Qwen2-VL-72B-Instruct-AWQ"
#MODEL_REPO = "Qwen/Qwen2-VL-7B-Instruct"
# Load the model and processor on available device(s)
model = Qwen2VLForConditionalGeneration.from_pretrained(
    MODEL_REPO,
    torch_dtype=torch.float16,
    #device_map="auto"
)#.to(device)

processor = AutoProcessor.from_pretrained(MODEL_REPO)

@spaces.GPU(duration=60)
def generate_caption(message, history, system_prompt, max_new_tokens):
    messages = [
        {
            "role": "user",
            "content": [
                {"type": "text", "text": message.get("text", "")}
            ]
        }
    ]
    for image in message["files"]:
        messages["content"].append({"type": "image", "image": image})  # The uploaded image
   
    # Prepare the input
    text = processor.apply_chat_template(
        messages, tokenize=False, add_generation_prompt=True
    )
    image_inputs, video_inputs = process_vision_info(messages)
    inputs = processor(
        text=[text],
        images=image_inputs,
        videos=video_inputs,
        padding=True,
        return_tensors="pt"
    )
    
    inputs.to(device)
    #model.to(device)

    # Generate the output
    generated_ids = model.generate(**inputs, max_new_tokens=max_new_tokens)
    generated_ids_trimmed = [
        out_ids[len(in_ids):] for in_ids, out_ids in zip(inputs.input_ids, generated_ids)
    ]
    output_text = processor.batch_decode(
        generated_ids_trimmed, skip_special_tokens=True, clean_up_tokenization_spaces=False
    )
    return output_text[0]

# Launch the Gradio interface with the updated inference function and title
with gr.Blocks() as demo:
    system_prompt = gr.Textbox("You are helpful AI.", label="System Prompt", render=False)
    tokens = gr.Slider(minimum=1, maximum=4096, value=128, step=1, label="Max new tokens", render=False)

    gr.ChatInterface(fn=generate_caption, title="Qwen2-VL-72B-Instruct-OCR", multimodal=True,
                    additional_inputs=[system_prompt, tokens],
                    description="Upload your Image and get the best possible insights out of the Image")

demo.queue().launch()