app.py

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("ibm-granite/granite-vision-3.1-2b-preview")

model = LlavaNextForConditionalGeneration.from_pretrained("ibm-granite/granite-vision-3.1-2b-preview", torch_dtype=torch.float16, low_cpu_mem_usage=True)
model.to("cuda:0")

@spaces.GPU
def bot_streaming(message, history):
  print(message)
  image = None
  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]

  if image is None:
      gr.Error("You need to upload an image for LLaVA to work.")
  prompt=f"[INST] <image>\n{message['text']} [/INST]"
  image = Image.open(image).convert("RGB")
  inputs = processor(image, prompt, return_tensors="pt").to("cuda:0")

  streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": True})
  generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024)
  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:
    print(f"{new_text = }")
      
    buffer += new_text
    
    generated_text_without_prompt = buffer[len(text_prompt):]
    time.sleep(0.04)
    print(f"{generated_text_without_prompt = }")
    yield generated_text_without_prompt


demo = gr.ChatInterface(fn=bot_streaming, title="Granite Vision 3.1 2b Preview (a LLaVA NeXT model)", 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/docs/transformers/main/en/model_doc/llava_next) in this demo (more specifically, the [Granite Vision-2B Preview variant](https://huggingface.co/ibm-granite/granite-vision-3.1-2b-preview)). Upload an image and start chatting about it, or simply try one of the examples below. If you don't upload an image, you will receive an error.",
                        stop_btn="Stop Generation", multimodal=True)
demo.launch(debug=True)