app.py

import json
import spaces
import subprocess
import gradio as gr
from llama_cpp import Llama
from huggingface_hub import hf_hub_download

CSS = """
#qwen-md .katex-display { display: inline; }
#qwen-md .katex-display>.katex { display: inline; }
#qwen-md .katex-display>.katex>.katex-html { display: inline; }
"""

hf_hub_download(
    repo_id="bartowski/Qwen2.5-Math-7B-Instruct-GGUF",
    filename="Qwen2.5-Math-7B-Instruct-Q6_K_L.gguf",
    local_dir="./models",
)

llm = Llama(
    model_path="models/Qwen2.5-Math-7B-Instruct-Q6_K_L.gguf",
    flash_attn=True,
    n_ctx=8192,
    n_batch=1024,
    chat_format="chatml",
)

# Gradio 组件
output_md = gr.Markdown(
    label="Answer",
    value="Answer will be presented here",
    latex_delimiters=[
        {"left": "\\(", "right": "\\)", "display": True},
        {"left": "\\begin\{equation\}", "right": "\\end\{equation\}", "display": True},
        {"left": "\\begin\{align\}", "right": "\\end\{align\}", "display": True},
        {"left": "\\begin\{alignat\}", "right": "\\end\{alignat\}", "display": True},
        {"left": "\\begin\{gather\}", "right": "\\end\{gather\}", "display": True},
        {"left": "\\begin\{CD\}", "right": "\\end\{CD\}", "display": True},
        {"left": "\\[", "right": "\\]", "display": True},
    ],
    elem_id="qwen-md",
    show_copy_button=True,
    container=True,
    render=False,
)
target_lang = gr.Dropdown(
    choices=["Chinese", "English"],
    value="Chinese",
    label="Output Language",
    interactive=True,
    render=False,
)
new_tokens = gr.Slider(
    minimum=1, maximum=8192, value=2048, step=1, label="Max new tokens", render=False
)
temperature = gr.Slider(
    minimum=0, maximum=2.0, value=0.5, step=0.1, label="Temperature", render=False
)
top_p = gr.Slider(minimum=0.0, maximum=1.0, value=0.95, step=0.05, label="Top P", render=False)
input_text = gr.Textbox(label="Ask math questions here", render=False)
submit_btn = gr.Button(value="Ask", render=False)
banner = gr.Markdown(value="""
# 📖 Qwen2.5-Math GGUF
This chatbot is based on the Qwen2.5-Math-7B-Instruct-GGUF model. You can ask it any math problem.
"""
)


# Gradio 函数
def respond(
    input_text,
    lang="Chinese",
    max_tokens=300,
    temperature=0.5,
    top_p=0.9,
):
    if lang == "Chinese":
        sys_msg = "你是一个乐于助人的数学助手. 你使用中文回答问题"
    else:
        sys_msg = "You are a helpful math assistant. You should always provide your answer in English."
    messages = [
        {
            "role": "system",
            "content": sys_msg,
        },
        {"role": "user", "content": input_text},
    ]

    response = ""
    response = llm.create_chat_completion(
        messages=messages,
        stream=True,
        max_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
    )
    message_repl = ""
    for chunk in response:
        if len(chunk['choices'][0]["delta"]) != 0 and "content" in chunk['choices'][0]["delta"]:
            message_repl = message_repl + \
                chunk['choices'][0]["delta"]["content"]
        yield message_repl


with gr.Blocks(css=CSS, theme="NoCrypt/miku") as demo:
    submit_btn.click(
        fn=respond,
        inputs=[input_text, target_lang, new_tokens, temperature, top_p],
        outputs=output_md,
    )
    with gr.Column():
        banner.render()
        with gr.Row():
            with gr.Column():
                input_text.render()
                target_lang.render()
                new_tokens.render()
                temperature.render()
                top_p.render()
                submit_btn.render()
            with gr.Column():
                output_md.render()

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