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: clean_text = re.sub("<$|