app.py

import spaces
import torch
import re
import gradio as gr
from threading import Thread
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM

import subprocess
subprocess.run('pip install flash-attn --no-build-isolation', env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, shell=True)

model_id = "vikhyatk/moondream2"
revision = "2024-04-02"
tokenizer = AutoTokenizer.from_pretrained(model_id, revision=revision)
moondream = AutoModelForCausalLM.from_pretrained(
    model_id, trust_remote_code=True, revision=revision,
    torch_dtype=torch.bfloat16, device_map={"": "cuda"},
    attn_implementation="flash_attention_2"
)
moondream.eval()


@spaces.GPU(duration=10)
def answer_question(img, prompt):
    image_embeds = moondream.encode_image(img)
    streamer = TextIteratorStreamer(tokenizer, skip_special_tokens=True)
    thread = Thread(
        target=moondream.answer_question,
        kwargs={
            "image_embeds": image_embeds,
            "question": prompt,
            "tokenizer": tokenizer,
            "streamer": streamer,
        },
    )
    thread.start()

    buffer = ""
    for new_text in streamer:
        buffer += new_text
        yield buffer.strip()


with gr.Blocks(theme="Monochrome") as demo:
    gr.Markdown(
        """
        # AskMoondream: Moondream 2 Demonstration Space
        Moondream2 is a 1.86B parameter model initialized with weights from SigLIP and Phi 1.5.
        Modularity AI presents this open source huggingface space for running fast experimental inferences on Moondream2.
        """
    )
    with gr.Row():
        prompt = gr.Textbox(label="Input", value="Describe this image.", scale=4)
        submit = gr.Button("Submit")
    with gr.Row():
        img = gr.Image(type="pil", label="Upload an Image")
        output = gr.TextArea(label="Response")
    submit.click(answer_question, [img, prompt], output)
    prompt.submit(answer_question, [img, prompt], output)

demo.queue().launch()