app.py

# app.py
import os
import io
import base64
from typing import List, Dict, Any, Optional

import httpx
import gradio as gr
from PIL import Image

# ========= 基本配置（可用环境变量覆写）=========
STEPFUN_ENDPOINT = os.getenv("STEPFUN_ENDPOINT", "https://api.stepfun.com/v1")
MODEL_NAME = os.getenv("STEPFUN_MODEL", "step-3")  # 可改为 step-r1-v-mini
REQUEST_TIMEOUT = float(os.getenv("REQUEST_TIMEOUT", "60"))
MAX_CHARS = int(os.getenv("MAX_CHARS", "20000"))   # 返回文本最大展示长度
# ===========================================


def _get_api_key() -> Optional[str]:
    """
    优先读 OPENAI_API_KEY（OpenAI 兼容习惯），否则读 STEPFUN_KEY。
    在 HF Space → Settings → Variables and secrets 添加其中一个即可。
    """
    return os.getenv("OPENAI_API_KEY") or os.getenv("STEPFUN_KEY")


def _pil_to_data_url(img: Image.Image, fmt: str = "PNG") -> str:
    """
    PIL.Image -> data:image/...;base64,... 字符串（适配 OpenAI 兼容的 image_url）
    """
    buf = io.BytesIO()
    img.save(buf, format=fmt)
    b64 = base64.b64encode(buf.getvalue()).decode("utf-8")
    mime = "image/png" if fmt.upper() == "PNG" else "image/jpeg"
    return f"data:{mime};base64,{b64}"


def _truncate(text: str, limit: int = MAX_CHARS) -> str:
    """
    软截断，避免一次性把超长内容写给前端导致传输异常。
    """
    if text is None:
        return ""
    if len(text) <= limit:
        return text
    return text[:limit] + "\n\n[输出过长，已截断]"


def _post_chat(messages: List[Dict[str, Any]], temperature: float = 0.7,
               max_tokens: Optional[int] = None) -> str:
    """
    直接请求 StepFun 的 /v1/chat/completions（OpenAI 兼容）。
    返回纯字符串，不抛异常，交由上层统一处理。
    """
    api_key = _get_api_key()
    if not api_key:
        # 不 raise，让 UI 只显示字符串，避免 Uvicorn/h11 生成异常页
        return ("[配置错误] 未检测到 API Key。\n"
                "请到 Space 的 Settings → Variables and secrets 添加：\n"
                "  OPENAI_API_KEY=你的 StepFun API Key   （或使用 STEPFUN_KEY）")

    url = f"{STEPFUN_ENDPOINT.rstrip('/')}/chat/completions"
    headers = {
        "Authorization": f"Bearer {api_key}",
        "Content-Type": "application/json",
    }
    payload: Dict[str, Any] = {
        "model": MODEL_NAME,
        "messages": messages,
        "temperature": temperature,
    }
    if max_tokens is not None:
        payload["max_tokens"] = max_tokens

    try:
        with httpx.Client(timeout=REQUEST_TIMEOUT) as client:
            resp = client.post(url, headers=headers, json=payload)
            resp.raise_for_status()
            data = resp.json()
        # 标准 OpenAI 兼容返回
        content = data["choices"][0]["message"]["content"]
        return _truncate(str(content))
    except httpx.HTTPStatusError as e:
        body = e.response.text if e.response is not None else repr(e)
        code = getattr(e.response, "status_code", "?")
        return _truncate(f"[HTTP {code}] 接口错误：\n{body}")
    except Exception as e:
        # 网络/解析/其他错误
        return _truncate(f"[调用失败] {repr(e)}")


def chat_with_step3(image: Optional[Image.Image], question: str, temperature: float) -> str:
    """
    Gradio 回调：接收图片与问题文本，返回字符串。
    任何异常都“吃掉”，只返回文本，防止框架层渲染异常页。
    """
    try:
        # 输入兜底
        if image is None and not question.strip():
            return "请上传一张图片，或至少输入一个问题。"

        content: List[Dict[str, Any]] = []
        if image is not None:
            data_url = _pil_to_data_url(image, fmt="PNG")
            content.append({"type": "image_url", "image_url": {"url": data_url}})

        if question.strip():
            content.append({"type": "text", "text": question.strip()})
        else:
            content.append({"type": "text", "text": "请描述这张图片。"})

        messages = [{"role": "user", "content": content}]
        return _post_chat(messages, temperature=temperature)
    except Exception as e:
        # 再兜一层底，避免任何未捕获异常冒泡
        return _truncate(f"[运行时错误] {repr(e)}")


# ================== Gradio UI ==================
with gr.Blocks(title="Step3 (StepFun API Demo)", analytics_enabled=False) as demo:
    gr.Markdown(
        """
        # Step3 · 图文对话演示（StepFun OpenAI 兼容接口）
        - 在 **Settings → Variables and secrets** 添加 `OPENAI_API_KEY`（或 `STEPFUN_KEY`）后即可使用  
        - 后端通过 `https://api.stepfun.com/v1/chat/completions`，不依赖 `openai` SDK  
        """
    )

    with gr.Row():
        image = gr.Image(type="pil", label="上传图片（可选）")

    question = gr.Textbox(label="问题", placeholder="例如：帮我看看这是什么菜，怎么做？")
    temperature = gr.Slider(0.0, 1.5, value=0.7, step=0.1, label="Temperature")
    submit = gr.Button("提交", variant="primary")
    output = gr.Textbox(label="模型回答", lines=12)

    submit.click(fn=chat_with_step3, inputs=[image, question, temperature], outputs=[output])

    gr.Markdown(
        """
        **小贴士：**
        - 如见到 `[配置错误] 未检测到 API Key`，请检查 Space 的 Secrets  
        - 如需改模型：设置环境变量 `STEPFUN_MODEL`，或在代码顶部修改默认值  
        - 如输出非常长，会自动做软截断避免传输异常  
        """
    )

if __name__ == "__main__":
    # 使用队列降低并发写冲突；关闭错误页，避免长 HTML 异常内容触发 h11 的 Content-Length 问题
    demo.queue(concurrency_count=2, max_size=32).launch(show_error=False, quiet=True)