chatbot_ui.py

#!/usr/bin/env python3
import os

import gradio as gr
from openai import OpenAI


VLLM_BASE_URL = os.getenv("VLLM_BASE_URL", "http://127.0.0.1:8000/v1")
VLLM_API_KEY = os.getenv("VLLM_API_KEY", "EMPTY")
DEFAULT_MODEL = os.getenv("VLLM_MODEL", "medgemma-4b-it")
CHATBOT_PORT = int(os.getenv("CHATBOT_PORT", "7860"))
ENABLE_SHARE = os.getenv("CHATBOT_SHARE", "false").strip().lower() in {
    "1",
    "true",
    "yes",
}
SYSTEM_PROMPT = os.getenv(
    "CHATBOT_SYSTEM_PROMPT",
    "You are a helpful medical assistant. Be clear, safe, and concise.",
)
GPT_STYLE_CSS = """
body, .gradio-container {
  background: #0b1020 !important;
  color: #e5e7eb !important;
}
#app-shell {
  max-width: 1400px !important;
  margin: 0 auto !important;
}
#left-sidebar {
  background: #111827 !important;
  border: 1px solid #1f2937 !important;
  border-radius: 16px !important;
  padding: 16px !important;
  min-height: 82vh !important;
}
#chat-panel {
  background: #0f172a !important;
  border: 1px solid #1f2937 !important;
  border-radius: 16px !important;
  padding: 8px 8px 0 8px !important;
  min-height: 82vh !important;
}
.gpt-title {
  font-size: 1.35rem;
  font-weight: 700;
  margin-bottom: 4px;
}
.gpt-subtitle {
  color: #94a3b8;
  margin-bottom: 12px;
}
.meta-chip {
  display: inline-block;
  font-size: 0.82rem;
  color: #cbd5e1;
  background: #1f2937;
  border: 1px solid #334155;
  border-radius: 999px;
  padding: 5px 10px;
  margin: 4px 8px 8px 0;
}
button.primary {
  background: #2563eb !important;
}
footer { display: none !important; }
"""


def get_client() -> OpenAI:
    return OpenAI(base_url=VLLM_BASE_URL, api_key=VLLM_API_KEY)


def normalize_content(content) -> str:
    if content is None:
        return ""
    if isinstance(content, str):
        return content
    if isinstance(content, list):
        parts = []
        for item in content:
            if isinstance(item, dict):
                if item.get("type") == "text" and item.get("text"):
                    parts.append(str(item["text"]))
                elif item.get("content"):
                    parts.append(str(item["content"]))
                else:
                    parts.append(str(item))
            else:
                parts.append(str(item))
        return "\n".join([p for p in parts if p]).strip()
    if isinstance(content, dict):
        if content.get("text"):
            return str(content["text"])
        if content.get("content"):
            return str(content["content"])
    return str(content)


def list_backend_models() -> list[str]:
    try:
        models = get_client().models.list().data
        return [m.id for m in models] or [DEFAULT_MODEL]
    except Exception:
        return [DEFAULT_MODEL]


AVAILABLE_MODELS = list_backend_models()
ACTIVE_MODEL = AVAILABLE_MODELS[0]


def chat_fn(message: str, history, model_id: str, temperature: float, max_tokens: int, system_prompt: str) -> str:
    messages = [{"role": "system", "content": system_prompt}]
    for item in history:
        if isinstance(item, (list, tuple)) and len(item) == 2:
            user_msg, assistant_msg = item
            if user_msg:
                messages.append({"role": "user", "content": normalize_content(user_msg)})
            if assistant_msg:
                messages.append({"role": "assistant", "content": normalize_content(assistant_msg)})
        elif isinstance(item, dict):
            role = item.get("role")
            content = item.get("content")
            if role in {"user", "assistant"} and content:
                messages.append({"role": role, "content": normalize_content(content)})
    messages.append({"role": "user", "content": normalize_content(message)})

    response = get_client().chat.completions.create(
        model=model_id,
        messages=messages,
        temperature=temperature,
        max_tokens=max_tokens,
    )
    return normalize_content(response.choices[0].message.content)


def main() -> None:
    with gr.Blocks(title="MedGemma Chatbot") as demo:
        with gr.Row(elem_id="app-shell"):
            with gr.Column(scale=3, elem_id="left-sidebar"):
                gr.HTML(
                    "<div class='gpt-title'>MedGemma Chat</div>"
                    "<div class='gpt-subtitle'>GPT-style dashboard for your vLLM backend.</div>"
                )
                gr.HTML(
                    f"<span class='meta-chip'>Active model: {ACTIVE_MODEL}</span>"
                    f"<span class='meta-chip'>Endpoint: {VLLM_BASE_URL}</span>"
                )
                gr.Markdown("### Settings")
                model_dd = gr.Dropdown(
                    choices=AVAILABLE_MODELS,
                    value=ACTIVE_MODEL if ACTIVE_MODEL in AVAILABLE_MODELS else AVAILABLE_MODELS[0],
                    label="Model",
                )
                temp_slider = gr.Slider(0.0, 1.2, value=0.2, step=0.05, label="Temperature")
                token_slider = gr.Slider(64, 2048, value=512, step=32, label="Max response tokens")
                prompt_box = gr.Textbox(
                    value=SYSTEM_PROMPT,
                    label="System prompt",
                    lines=4,
                )
                gr.Markdown(
                    "### Prompt ideas\n"
                    "- Summarize this clinical note in simple language.\n"
                    "- List red flags that need emergency care.\n"
                    "- Explain this diagnosis for a patient handout."
                )
            with gr.Column(scale=9, elem_id="chat-panel"):
                gr.ChatInterface(
                    fn=chat_fn,
                    chatbot=gr.Chatbot(height=690),
                    additional_inputs=[model_dd, temp_slider, token_slider, prompt_box],
                )

    demo.queue(default_concurrency_limit=32).launch(
        server_name="0.0.0.0",
        server_port=CHATBOT_PORT,
        share=ENABLE_SHARE,
        css=GPT_STYLE_CSS,
        show_error=True,
    )


if __name__ == "__main__":
    main()