fix(providers): avoid chat template body for vllm mistral

app-instance/backend/beaver/engine/providers/litellm.py

@@ -3,9 +3,11 @@
 from __future__ import annotations
 
 from contextlib import contextmanager
+from ipaddress import ip_address
 import json
 import os
 from typing import Any
+from urllib.parse import urlsplit
 
 from .base import LLMProvider, LLMResponse, ToolCallRequest
 from .registry import find_by_model, find_by_name, find_gateway
@@ -26,6 +28,23 @@ except ModuleNotFoundError:  # pragma: no cover
 _ALLOWED_MSG_KEYS = frozenset({"role", "content", "tool_calls", "tool_call_id", "name", "reasoning_content"})
 
 
+def _looks_like_local_vllm_api_base(api_base: str | None) -> bool:
+    if not api_base:
+        return False
+    lowered = api_base.lower()
+    if "vllm" in lowered or "localhost" in lowered:
+        return True
+
+    host = urlsplit(lowered).hostname or ""
+    if host in {"127.0.0.1", "::1", "0.0.0.0"}:
+        return True
+    try:
+        parsed_host = ip_address(host)
+    except ValueError:
+        return False
+    return parsed_host.is_private or parsed_host.is_loopback
+
+
 class LiteLLMProvider(LLMProvider):
     """通过 LiteLLM 统一访问大多数 provider。"""
 
@@ -200,10 +219,12 @@ class LiteLLMProvider(LLMProvider):
         kwargs["extra_body"] = extra_body
 
     def _uses_mistral_reasoning_parser(self, original_model: str, resolved_model: str) -> bool:
-        if self.provider_name != "vllm":
-            return False
         model_names = f"{original_model} {resolved_model}".lower()
-        return "mistral" in model_names
+        if "mistral" not in model_names:
+            return False
+        if self.provider_name == "vllm":
+            return True
+        return self.provider_name in {"openai", "custom"} and _looks_like_local_vllm_api_base(self.api_base)
 
     async def chat(
         self,