Spaces:
Paused
Paused
File size: 1,806 Bytes
002ed37 6174149 002ed37 |
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 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 |
import spaces
import argparse
import torch
import re
import gradio as gr
from threading import Thread
from transformers import TextIteratorStreamer, AutoTokenizer, AutoModelForCausalLM
parser = argparse.ArgumentParser()
if torch.cuda.is_available():
device, dtype = "cuda", torch.float16
else:
device, dtype = "cpu", torch.float32
model_id = "vikhyatk/moondream2"
tokenizer = AutoTokenizer.from_pretrained(model_id, revision="2024-03-06")
moondream = AutoModelForCausalLM.from_pretrained(
model_id, trust_remote_code=True, revision="2024-03-06"
).to(device=device, dtype=dtype)
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:
clean_text = re.sub("<$|<END$", "", new_text)
buffer += clean_text
yield buffer
with gr.Blocks() as demo:
gr.Image("data/redhen.ico")
gr.Markdown(
"""
# Super Rapid Annotator - Multimodal vision tool to annotate videos with LLaVA framework
"""
)
with gr.Row():
prompt = gr.Textbox(label="Input", placeholder="Type here...", 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()
|