app.py

import os
from threading import Thread
from typing import Iterator

import gradio as gr
import spaces
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))

DESCRIPTION = """\
# L-MChat
This Space demonstrates [L-MChat](https://huggingface.co/collections/Artples/l-mchat-663265a8351231c428318a8f) by L-AI.
"""

if not torch.cuda.is_available():
    DESCRIPTION += "\n<p>Running on CPU! This demo does not work on CPU.</p>"

model_options = {
    "Fast-Model": "Artples/L-MChat-Small",
    "Quality-Model": "Artples/L-MChat-7b"
}

@spaces.GPU(enable_queue=True, duration=90)
def generate(
    message: str,
    model_choice: str,
    chat_history: list[tuple[str, str]],
    system_prompt: str,
    max_new_tokens: int = 1024,
    temperature: float = 0.1,
    top_p: float = 0.9,
    top_k: int = 50,
    repetition_penalty: float = 1.2,
) -> Iterator[str]:
    model_id = model_options[model_choice]
    model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")
    tokenizer = AutoTokenizer.from_pretrained(model_id)
    tokenizer.use_default_system_prompt = False

    conversation = []
    if system_prompt:
        conversation.append({"role": "system", "content": system_prompt})
    for user, assistant in chat_history:
        conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
    conversation.append({"role": "user", "content": message})

    input_ids = tokenizer(conversation, return_tensors="pt", padding=True, truncation=True)
    if input_ids['input_ids'].shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids['input_ids'] = input_ids['input_ids'][:, -MAX_INPUT_TOKEN_LENGTH:]

    outputs = model.generate(
        **input_ids,
        max_length=input_ids['input_ids'].shape[1] + max_new_tokens,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_return_sequences=1,
        repetition_penalty=repetition_penalty
    )

    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
    yield generated_text

chat_interface = gr.Interface(
    fn=generate,
    inputs=[
        gr.Textbox(lines=2, placeholder="Type your message here..."),
        gr.Dropdown(label="Choose Model", choices=list(model_options.keys())),
        gr.State(label="Chat History", default=[]),
        gr.Textbox(label="System Prompt", lines=6, placeholder="Enter system prompt if any..."),
        gr.Slider(label="Max new tokens", minimum=1, maximum=MAX_MAX_NEW_TOKENS, step=1, value=DEFAULT_MAX_NEW_TOKENS),
        gr.Slider(label="Temperature", minimum=0.1, maximum=4.0, step=0.1, value=0.1),
        gr.Slider(label="Top-p (nucleus sampling)", minimum=0.05, maximum=1.0, step=0.05, value=0.9),
        gr.Slider(label="Top-k", minimum=1, maximum=1000, step=1, value=50),
        gr.Slider(label="Repetition penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.2),
    ],
    outputs=[gr.Textbox(label="Response")],
    theme="default",
    description=DESCRIPTION
)

if __name__ == "__main__":
    chat_interface.launch()