app.py

#!/usr/bin/env python

import os
from threading import Thread
from typing import Iterator

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

DESCRIPTION = "# CALM2-7B-chat"

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

MAX_MAX_NEW_TOKENS = 2048
DEFAULT_MAX_NEW_TOKENS = 1024
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "32768"))

if torch.cuda.is_available():
    model_id = "cyberagent/calm2-7b-chat"
    model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")
    tokenizer = AutoTokenizer.from_pretrained(model_id)


def apply_chat_template(conversation: list[dict[str, str]]) -> str:
    prompt = "\n".join([f"{c['role']}: {c['content']}" for c in conversation])
    prompt = f"{prompt}\nASSISTANT: "
    return prompt


@spaces.GPU
@torch.inference_mode()
def generate(
    message: str,
    chat_history: list[tuple[str, str]],
    max_new_tokens: int = 1024,
    temperature: float = 0.7,
    top_p: float = 0.95,
    top_k: int = 50,
    repetition_penalty: float = 1.0,
) -> Iterator[str]:
    conversation = []
    for user, assistant in chat_history:
        conversation.extend([{"role": "USER", "content": user}, {"role": "ASSISTANT", "content": assistant}])
    conversation.append({"role": "USER", "content": message})

    prompt = apply_chat_template(conversation)
    input_ids = tokenizer.encode(prompt, add_special_tokens=False, return_tensors="pt")
    if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
        input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
        gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
    input_ids = input_ids.to(model.device)

    streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
    generate_kwargs = dict(
        {"input_ids": input_ids},
        streamer=streamer,
        max_new_tokens=max_new_tokens,
        do_sample=True,
        top_p=top_p,
        top_k=top_k,
        temperature=temperature,
        num_beams=1,
        repetition_penalty=repetition_penalty,
    )
    t = Thread(target=model.generate, kwargs=generate_kwargs)
    t.start()

    outputs = []
    for text in streamer:
        outputs.append(text)
        yield "".join(outputs)


chat_interface = gr.ChatInterface(
    fn=generate,
    chatbot=gr.Chatbot(show_label=False, layout="panel", height=600),
    additional_inputs_accordion_name="詳細設定",
    additional_inputs=[
        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=0.7,
        ),
        gr.Slider(
            label="Top-p (nucleus sampling)",
            minimum=0.05,
            maximum=1.0,
            step=0.05,
            value=0.95,
        ),
        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=1.0,
        ),
    ],
    stop_btn=None,
    examples=[
        ["東京の観光名所を教えて。"],
        ["落武者って何？"],
        ["暴れん坊将軍って誰のこと？"],
        ["人がヘリを食べるのにかかる時間は？"],
    ],
)

with gr.Blocks(css="style.css") as demo:
    gr.Markdown(DESCRIPTION)
    gr.DuplicateButton(
        value="Duplicate Space for private use",
        elem_id="duplicate-button",
        visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
    )
    chat_interface.render()

if __name__ == "__main__":
    demo.queue(max_size=20).launch()