app.py

# import subprocess
# subprocess.run(
#     'pip install flash-attn --no-build-isolation', 
#     env={'FLASH_ATTENTION_SKIP_CUDA_BUILD': "TRUE"}, 
#     shell=True
# )
from threading import Thread
import torch
from PIL import Image
import gradio as gr
import spaces
from transformers import AutoModel, AutoTokenizer, TextIteratorStreamer
import os
import time


os.environ["HF_HUB_ENABLE_HF_TRANSFER"] = "1"

HF_TOKEN = os.environ.get("HF_TOKEN", None)
MODEL_ID = os.environ.get("MODEL_ID")
MODEL_NAME = MODEL_ID.split("/")[-1]

TITLE = "<h1><center>VL-Chatbox</center></h1>"

DESCRIPTION = f'<h3><center>MODEL: <a href="https://hf.co/{MODEL_ID}">{MODEL_NAME}</a></center></h3>'

CSS = """
.duplicate-button {
  margin: auto !important;
  color: white !important;
  background: black !important;
  border-radius: 100vh !important;
}
"""

model = AutoModel.from_pretrained(
    MODEL_ID,
    torch_dtype=torch.float16,
    trust_remote_code=True
).to(0)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
model.eval()



@spaces.GPU(queue=False)
def stream_chat(message, history: list, temperature: float, max_new_tokens: int):
    print(f'message is - {message}')
    print(f'history is - {history}')
    conversation = [] 
    if message["files"]:
        image = Image.open(message["files"][-1]).convert('RGB')
        conversation.append({"role": "user", "content": message['text']})
    else:
        if len(history) == 0:
            raise gr.Error("Please upload an image first.")
            image = None
        else:
            image = Image.open(history[0][0][0])
            for prompt, answer in history:
                conversation.extend([{"role": "user", "content": prompt}, {"role": "assistant", "content": answer}])
            conversation.append({"role": "user", "content": message['text']})    
    print(f"Conversation is -\n{conversation}")

    generate_kwargs = dict(
        image=image,
        msgs=conversation,
        max_new_tokens=max_new_tokens,
        temperature=temperature,
        sampling=True,
        tokenizer=tokenizer,
    )
    if temperature == 0:
        generate_kwargs["sampling"] = False

    response = model.chat(**generate_kwargs)
    return response


chatbot = gr.Chatbot(height=450)
chat_input = gr.MultimodalTextbox(
    interactive=True, 
    file_types=["image"], 
    placeholder="Enter message or upload file...", 
    show_label=False,

)
EXAMPLES = [
        [{"text": "What is on the desk?", "files": ["./laptop.jpg"]}],
        [{"text": "Where it is?", "files": ["./hotel.jpg"]}],
        [{"text": "Can yo describe this image?", "files": ["./spacecat.png"]}]
]

with gr.Blocks(css=CSS) as demo:
    gr.HTML(TITLE)
    gr.HTML(DESCRIPTION)
    gr.DuplicateButton(value="Duplicate Space for private use", elem_classes="duplicate-button")
    gr.ChatInterface(
        fn=stream_chat,
        multimodal=True,
        textbox=chat_input,
        chatbot=chatbot,
        fill_height=True,
        additional_inputs_accordion=gr.Accordion(label="⚙️ Parameters", open=False, render=False),
        additional_inputs=[
            gr.Slider(
                minimum=0,
                maximum=1,
                step=0.1,
                value=0.8,
                label="Temperature",
                render=False,
            ),
            gr.Slider(
                minimum=128,
                maximum=4096,
                step=1,
                value=1024,
                label="Max new tokens",
                render=False,
            ),
        ],
    ),
    gr.Examples(EXAMPLES,[chat_input])


if __name__ == "__main__":
    demo.queue(api_open=False).launch(show_api=False, share=False)