File size: 1,500 Bytes
eb99994 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 |
import spaces
import argparse
import torch
import re
import gradio as gr
from threading import Thread
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
from PIL import Image
parser = argparse.ArgumentParser()
model_id = "vikhyat/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.float32
)
moondream.eval()
@spaces.GPU(duration=10)
def answer_question(images, prompts):
image_embeds = [moondream.encode_image(img) for img in images]
image_embeds = torch.cat(image_embeds, dim=0)
answers = moondream.batch_answer(
images=image_embeds,
prompts=prompts,
tokenizer=tokenizer
)
return [answer for answer in answers]
with gr.Blocks() as demo:
gr.Markdown(
"""
# 🌔 moondream2
A tiny vision language model. [GitHub](https://github.com/vikhyat/moondream)
"""
)
with gr.Row():
prompts = gr.Textbox(label="Input", placeholder="Type here...", scale=4)
submit = gr.Button("Submit")
with gr.Row():
images = gr.Image(type="pil", label="Upload Images", multiple=True)
output = gr.Textbox(label="Response", multiple=True)
submit.click(answer_question, [images, prompts], output)
prompts.submit(answer_question, [images, prompts], output)
demo.queue().launch() |