File size: 2,070 Bytes
f04732f |
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 |
import gradio as gr
from transformers import LlavaNextProcessor, LlavaNextForConditionalGeneration, TextIteratorStreamer
from threading import Thread
import re
import time
from PIL import Image
import torch
import spaces
processor = LlavaNextProcessor.from_pretrained("llava-hf/llava-v1.6-mistral-7b-hf")
model = LlavaNextForConditionalGeneration.from_pretrained("llava-hf/llava-v1.6-mistral-7b-hf", torch_dtype=torch.float16, low_cpu_mem_usage=True)
model.to("cuda:0")
@spaces.GPU
def bot_streaming(message, history):
print(message)
if message["files"]:
image = message["files"][-1]["path"]
else:
# if there's no image uploaded for this turn, look for images in the past turns
# kept inside tuples, take the last one
for hist in history:
if type(hist[0])==tuple:
image = hist[0][0]
prompt=f"[INST] <image>\n{message['text']} [/INST]"
image = Image.open(image).convert("RGB")
inputs = processor(prompt, image, return_tensors="pt").to("cuda:0")
streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": True})
generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=100)
generated_text = ""
thread = Thread(target=model.generate, kwargs=generation_kwargs)
thread.start()
text_prompt =f"[INST] \n{message['text']} [/INST]"
buffer = ""
for new_text in streamer:
buffer += new_text
generated_text_without_prompt = buffer[len(text_prompt):]
time.sleep(0.04)
yield generated_text_without_prompt
demo = gr.ChatInterface(fn=bot_streaming, title="LLaVA Next", examples=[{"text": "What is on the flower?", "files":["./bee.jpg"]},
{"text": "How to make this pastry?", "files":["./baklava.png"]}],
description="Try [LLaVA Next](https://huggingface.co/papers/2310.03744) in this demo. Upload an image and start chatting about it, or simply try one of the examples below.",
stop_btn="Stop Generation", multimodal=True)
demo.launch(debug=True) |