增加openai

2026-05-22 22:31:09 +08:00
parent ac762b540c
commit 5ff9cc4587
5 changed files with 202 additions and 42 deletions
@@ -119,12 +119,14 @@ class MonitorConfig(BaseModel):

 class GemmaConfig(BaseModel):
    """
-    本地 Ollama 跑 Gemma（或其它模型）做漏斗二次分拣。
-    需在机器上自行启动 ollama 并拉取模型；开启后仅对本轮 5m 扫描命中的 WATCH/TRIGGER 按成交额取前 N 再请求。
+    Gemma 漏斗：默认直连本机 Ollama（/api/chat）。
+    若使用 OpenAI 兼容网关（如 https://op.bz121.com/v1 + Bearer），设 api_style=openai 并填写 api_key。
    """

    enabled: bool = False
    ollama_base_url: str = "http://127.0.0.1:11434"
+    api_key: str = ""
+    api_style: Literal["ollama", "openai"] = "ollama"
    model: str = "gemma2:2b"
    timeout_seconds: float = 180.0
    temperature: float = 0.15
@@ -3,7 +3,7 @@ from __future__ import annotations
 import json
 import logging
 import re
-from typing import Any
+from typing import Any, Literal

 import httpx

@@ -24,11 +24,108 @@ def _extract_json_object(text: str) -> dict[str, Any] | None:
        return None


+def _resolve_api_style(conf: GemmaConfig) -> Literal["ollama", "openai"]:
+    style = (conf.api_style or "ollama").strip().lower()
+    if style == "openai":
+        return "openai"
+    if style == "ollama":
+        return "ollama"
+    if (conf.api_key or "").strip():
+        return "openai"
+    return "ollama"
+
+
+def _chat_url(conf: GemmaConfig, style: Literal["ollama", "openai"]) -> str:
+    base = conf.ollama_base_url.rstrip("/")
+    if style == "openai":
+        if base.endswith("/v1"):
+            return f"{base}/chat/completions"
+        return f"{base}/v1/chat/completions"
+    return f"{base}/api/chat"
+
+
+def _build_headers(conf: GemmaConfig, style: Literal["ollama", "openai"]) -> dict[str, str]:
+    headers = {"Content-Type": "application/json"}
+    if style == "openai":
+        key = (conf.api_key or "").strip()
+        if key:
+            headers["Authorization"] = f"Bearer {key}"
+    return headers
+
+
+def _strip_b64_prefix(raw: str) -> str:
+    s = raw.strip()
+    if "," in s and s.lower().startswith("data:"):
+        return s.split(",", 1)[1]
+    return s
+
+
 class OllamaGemmaClient:
    def __init__(self, conf: GemmaConfig) -> None:
        self.conf = conf
+        self._style = _resolve_api_style(conf)
        self.timeout = httpx.Timeout(conf.timeout_seconds, read=conf.timeout_seconds + 30.0)

+    async def _chat(
+        self,
+        *,
+        system: str,
+        user_text: str,
+        image_base64: str | None,
+        temperature: float,
+        json_mode: bool,
+    ) -> str:
+        url = _chat_url(self.conf, self._style)
+        headers = _build_headers(self.conf, self._style)
+
+        if self._style == "openai":
+            user_content: str | list[dict[str, Any]] = user_text
+            if image_base64 and self.conf.send_chart_image:
+                b64 = _strip_b64_prefix(image_base64)
+                user_content = [
+                    {"type": "text", "text": user_text},
+                    {
+                        "type": "image_url",
+                        "image_url": {"url": f"data:image/png;base64,{b64}"},
+                    },
+                ]
+            payload: dict[str, Any] = {
+                "model": self.conf.model,
+                "messages": [
+                    {"role": "system", "content": system},
+                    {"role": "user", "content": user_content},
+                ],
+                "temperature": temperature,
+                "stream": False,
+            }
+            if json_mode:
+                payload["response_format"] = {"type": "json_object"}
+        else:
+            message: dict[str, Any] = {"role": "user", "content": user_text}
+            if image_base64 and self.conf.send_chart_image:
+                message["images"] = [_strip_b64_prefix(image_base64)]
+            payload = {
+                "model": self.conf.model,
+                "messages": [{"role": "system", "content": system}, message],
+                "stream": False,
+                "options": {"temperature": temperature},
+            }
+            if json_mode:
+                payload["format"] = "json"
+
+        async with httpx.AsyncClient(timeout=self.timeout, trust_env=False) as client:
+            resp = await client.post(url, json=payload, headers=headers)
+            resp.raise_for_status()
+            data = resp.json()
+
+        if self._style == "openai":
+            choices = data.get("choices") or []
+            if choices and isinstance(choices[0], dict):
+                msg = (choices[0].get("message") or {}).get("content") or ""
+                return str(msg)
+            return ""
+        return str((data.get("message") or {}).get("content") or "")
+
    async def rank_funnel(
        self,
        symbol: str,
@@ -36,9 +133,7 @@ class OllamaGemmaClient:
        ohlc_csv_block: str,
        image_base64: str | None,
    ) -> dict[str, Any]:
-        """
-        调用本地 Ollama，让 Gemma 按漏斗标准 JSON 回复。
-        """
+        """调用 Ollama 或 OpenAI 兼容网关，让 Gemma 按漏斗标准 JSON 回复。"""
        system = (
            "你是加密货币永续合约的日线结构分析师。只输出一个 JSON 对象，不要 Markdown，不要代码围栏。"
            "字段必须全部存在且为英文枚举/数字："
@@ -55,29 +150,44 @@ class OllamaGemmaClient:
            f"程序化摘要:\n{programmatic_text}\n\n"
            f"最近日线 OHLCV（时间正序最后一行为最新）:\n{ohlc_csv_block}\n"
        )
-        url = f"{self.conf.ollama_base_url.rstrip('/')}/api/chat"
-        message: dict[str, Any] = {"role": "user", "content": user_body}
-        if image_base64 and self.conf.send_chart_image:
-            message["images"] = [image_base64]
+        try:
+            msg = await self._chat(
+                system=system,
+                user_text=user_body,
+                image_base64=image_base64,
+                temperature=self.conf.temperature,
+                json_mode=self.conf.json_mode,
+            )
+        except httpx.HTTPStatusError as exc:
+            LOGGER.warning(
+                "gemma_http_error symbol=%s status=%s url=%s",
+                symbol,
+                exc.response.status_code,
+                exc.request.url,
+            )
+            return {
+                "error": f"http_{exc.response.status_code}",
+                "raw": (exc.response.text or "")[:500],
+                "daily_structure": "weak",
+                "volume_view": "low",
+                "upside_space": "low",
+                "mid_resistance": "high",
+                "priority": 1,
+                "one_liner": f"模型网关 HTTP {exc.response.status_code}",
+            }
+        except Exception as exc:  # noqa: BLE001
+            LOGGER.warning("gemma_request_failed symbol=%s: %s", symbol, exc)
+            return {
+                "error": str(exc),
+                "daily_structure": "weak",
+                "volume_view": "low",
+                "upside_space": "low",
+                "mid_resistance": "high",
+                "priority": 1,
+                "one_liner": f"模型调用失败: {exc}",
+            }

-        payload: dict[str, Any] = {
-            "model": self.conf.model,
-            "messages": [{"role": "system", "content": system}, message],
-            "stream": False,
-            "options": {"temperature": self.conf.temperature},
-        }
-        if self.conf.json_mode:
-            payload["format"] = "json"
-
-        async with httpx.AsyncClient(timeout=self.timeout, trust_env=False) as client:
-            resp = await client.post(url, json=payload)
-            resp.raise_for_status()
-            data = resp.json()
-
-        msg = (data.get("message") or {}).get("content") or ""
        parsed = _extract_json_object(msg) if msg else None
-        if parsed is None and isinstance(data.get("message"), dict):
-            parsed = _extract_json_object(str(data["message"]))
        if parsed is None:
            LOGGER.warning("gemma_parse_failed symbol=%s raw_len=%s", symbol, len(msg))
            return {
@@ -105,20 +215,17 @@ class OllamaGemmaClient:
            "要求：1) headline 一句话；2) btc_explain 解释方向；"
            "3) summary 覆盖 WATCH/TRIGGER/漏斗；4) risk_points 给1-3条；5) action_hint 给执行提示。"
        )
-        url = f"{self.conf.ollama_base_url.rstrip('/')}/api/chat"
-        payload: dict[str, Any] = {
-            "model": self.conf.model,
-            "messages": [{"role": "system", "content": system}, {"role": "user", "content": user_body}],
-            "stream": False,
-            "options": {"temperature": 0.1},
-            "format": "json",
-        }
-        async with httpx.AsyncClient(timeout=self.timeout, trust_env=False) as client:
-            resp = await client.post(url, json=payload)
-            resp.raise_for_status()
-            data = resp.json()
+        try:
+            msg = await self._chat(
+                system=system,
+                user_text=user_body,
+                image_base64=None,
+                temperature=0.1,
+                json_mode=True,
+            )
+        except Exception as exc:  # noqa: BLE001
+            return {"error": "parse_failed", "raw": str(exc)[:1200]}

-        msg = (data.get("message") or {}).get("content") or ""
        parsed = _extract_json_object(msg) if msg else None
        if parsed is None:
            return {"error": "parse_failed", "raw": msg[:1200]}
@@ -724,6 +724,8 @@ def create_app(settings: Settings) -> FastAPI:
                "gemma": {
                    "enabled": g.enabled,
                    "ollama_base_url": g.ollama_base_url,
+                    "api_style": g.api_style,
+                    "api_key_set": bool((g.api_key or "").strip()),
                    "model": g.model,
                    "max_funnel_per_cycle": g.max_funnel_per_cycle,
                    "vision_top_n": g.vision_top_n,