app.py

import time
from threading import Thread

import gradio as gr
import torch
from PIL import Image
from transformers import AutoProcessor, LlavaForConditionalGeneration
from transformers import TextIteratorStreamer

import spaces


PLACEHOLDER = """
<div style="padding: 30px; text-align: center; display: flex; flex-direction: column; align-items: center;">
   <img src="https://cdn-uploads.huggingface.co/production/uploads/64ccdc322e592905f922a06e/DDIW0kbWmdOQWwy4XMhwX.png" style="width: 80%; max-width: 550px; height: auto; opacity: 0.55;  "> 
   <h1 style="font-size: 28px; margin-bottom: 2px; opacity: 0.55;">LLaVA-Llama-3-8B</h1>
   <p style="font-size: 18px; margin-bottom: 2px; opacity: 0.65;">Llava-Llama-3-8b is a LLaVA model fine-tuned from Meta-Llama-3-8B-Instruct and CLIP-ViT-Large-patch14-336 with ShareGPT4V-PT and InternVL-SFT by XTuner</p>
</div>
"""


model_id = "xtuner/llava-llama-3-8b-v1_1-transformers"

processor = AutoProcessor.from_pretrained(model_id)

model = LlavaForConditionalGeneration.from_pretrained(
    model_id,
    torch_dtype=torch.float16,
    low_cpu_mem_usage=True,
)

model.to("cuda:0")
model.generation_config.eos_token_id = 128009


@spaces.GPU
def bot_streaming(message, history):
    print(message)
    if message["files"]:
        # message["files"][-1] is a Dict or just a string
        if type(message["files"][-1]) == dict:
            image = message["files"][-1]["path"]
        else:
            image = message["files"][-1]
    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:
        # Handle the case where image is None
        image = "http://files.tatl.tv/ignore.png"

    prompt = f"<|start_header_id|>user<|end_header_id|>\n\n<image>\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
    # print(f"prompt: {prompt}")
    image = Image.open(image)
    inputs = processor(prompt, image, return_tensors='pt').to(0, torch.float16)

    streamer = TextIteratorStreamer(processor, **{"skip_special_tokens": False, "skip_prompt": True})
    generation_kwargs = dict(inputs, streamer=streamer, max_new_tokens=1024, do_sample=False)

    thread = Thread(target=model.generate, kwargs=generation_kwargs)
    thread.start()

    text_prompt = f"<|start_header_id|>user<|end_header_id|>\n\n{message['text']}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n"
    # print(f"text_prompt: {text_prompt}")

    buffer = ""
    time.sleep(0.5)
    for new_text in streamer:
        # find <|eot_id|> and remove it from the new_text
        if "<|eot_id|>" in new_text:
            new_text = new_text.split("<|eot_id|>")[0]
        buffer += new_text

        # generated_text_without_prompt = buffer[len(text_prompt):]
        generated_text_without_prompt = buffer
        # print(generated_text_without_prompt)
        time.sleep(0.06)
        # print(f"new_text: {generated_text_without_prompt}")
        yield generated_text_without_prompt


chatbot=gr.Chatbot(placeholder=PLACEHOLDER,scale=1)
chat_input = gr.MultimodalTextbox(interactive=True, file_types=["image"], placeholder="Enter message or upload file...", show_label=False)
with gr.Blocks(fill_height=True, ) as demo:
    gr.ChatInterface(
    fn=bot_streaming,
    description="Try [LLaVA Llama-3-8B](https://huggingface.co/xtuner/llava-llama-3-8b-v1_1-transformers).",
    stop_btn="Stop Generation",
    multimodal=True,
    textbox=chat_input,
    chatbot=chatbot,
    )

demo.queue(api_open=False)
demo.launch(show_api=False, share=False)