import os
import gradio as gr
from openai import OpenAI

title = "ERNIE4.5-21B-A3B-Thinking Demo"
description = """
- Official Website: <https://yiyan.baidu.com/> (UI in Chinese)
- API services: [Qianfan Large Model Platform](https://cloud.baidu.com/product-s/qianfan_home) (cloud platform providing LLM services, UI in Chinese)
- Open weights on Hugging Face: [ERNIE4.5-21B-A3B-Thinking](https://huggingface.co/baidu/ERNIE-4.5-21B-A3B-Thinking) 
"""

qianfan_api_key = os.getenv("QIANFAN_TOKEN")
qianfan_model = "ernie-4.5-21b-a3b-thinking"


client = OpenAI(base_url="https://qianfan.baidubce.com/v2", api_key=qianfan_api_key)


def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
):
    messages = [{"role": "system", "content": system_message}]
    messages.extend(history)
    messages.append({"role": "user", "content": message})

    response = client.chat.completions.create(
        model=qianfan_model,
        messages=messages,
        max_completion_tokens=max_tokens,
        stream=True,
    )

    reasoning_content = "**Thinking**:\n"
    content = "\n\n**Answer**: \n"

    for chunk in response:
        if hasattr(chunk.choices[0].delta, 'reasoning_content'):
            token = chunk.choices[0].delta.reasoning_content
            if token:
                reasoning_content += token
                yield reasoning_content
        elif hasattr(chunk.choices[0].delta, 'content'):
            token = chunk.choices[0].delta.content
            if token:
                content += token
                yield reasoning_content + content


demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="", label="System message"),
        gr.Slider(minimum=1024, maximum=65536, value=32768, step=1024, label="Max new tokens"),
    ],
    title=title,
    description=description,
    type='messages',
    concurrency_limit=50
)

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