From 0d6a82ba142d8670e1ae9e4cf46b6c3da418c2ba Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E6=B5=85=E5=BF=B5?= <email@sample.com>
Date: Tue, 20 Oct 2026 17:23:36 +0800
Subject: [PATCH] =?UTF-8?q?feat(glm):=20max=5Ftokens=20=E5=8F=AF=E9=80=89?=
 =?UTF-8?q?=E9=80=8F=E4=BC=A0=EF=BC=88site-import=20=E5=89=8D=E7=BD=AE?=
 =?UTF-8?q?=EF=BC=89?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

max_tokens 探针结论: 8192 可用（glm-4-flash 实测 max_tokens=8192 调用成功返回，finish_reason=stop；Task 5 可按 8192 取值）

Co-Authored-By: Claude Code <noreply@anthropic.com>
---
 app/engine/providers/glm.py | 10 +++++++++-
 tests/test_glm_provider.py  | 34 ++++++++++++++++++++++++++++++++++
 2 files changed, 43 insertions(+), 1 deletion(-)
 create mode 100644 tests/test_glm_provider.py

diff --git a/app/engine/providers/glm.py b/app/engine/providers/glm.py
index 78b328b..9b9f9ed 100644
--- a/app/engine/providers/glm.py
+++ b/app/engine/providers/glm.py
@@ -13,9 +13,16 @@ DEFAULT_MODEL = "glm-4-flash"
 
 
 class GLMProvider(AIProvider):
-    def __init__(self, api_key: str, model: str = "glm-4-flash", base_url: str = ""):
+    def __init__(
+        self,
+        api_key: str,
+        model: str = "glm-4-flash",
+        base_url: str = "",
+        max_tokens: int | None = None,
+    ):
         self.client = ZhipuAI(api_key=api_key, base_url=base_url or None)
         self.model = model
+        self.max_tokens = max_tokens
 
     async def chat(self, messages: list[dict], system_prompt: str) -> str:
         full_messages = [{"role": "system", "content": system_prompt}] + messages
@@ -74,6 +81,7 @@ class GLMProvider(AIProvider):
                 model=self.model,
                 messages=messages,
                 temperature=temperature,
+                **({"max_tokens": self.max_tokens} if self.max_tokens is not None else {}),
             )
             return resp.choices[0].message.content
 
diff --git a/tests/test_glm_provider.py b/tests/test_glm_provider.py
new file mode 100644
index 0000000..f5afe82
--- /dev/null
+++ b/tests/test_glm_provider.py
@@ -0,0 +1,34 @@
+"""GLMProvider max_tokens 透传（site-import spec §3.2）。"""
+from unittest.mock import MagicMock, patch
+
+from app.engine.providers.glm import GLMProvider
+
+
+def _provider(**kw):
+    with patch("app.engine.providers.glm.ZhipuAI"):
+        return GLMProvider(api_key="k", **kw)
+
+
+def _ok(cli):
+    cli.chat.completions.create.return_value.choices = [
+        MagicMock(message=MagicMock(content="ok"))]
+
+
+class TestMaxTokensPassthrough:
+    def test_默认不传_调用无max_tokens参数(self):
+        import asyncio
+        p = _provider()
+        with patch.object(p, "client") as cli:
+            _ok(cli)
+            asyncio.run(p.chat([{"role": "user", "content": "hi"}], "sys"))
+            _, kwargs = cli.chat.completions.create.call_args
+            assert "max_tokens" not in kwargs
+
+    def test_传入max_tokens_透传到create(self):
+        import asyncio
+        p = _provider(max_tokens=4096)
+        with patch.object(p, "client") as cli:
+            _ok(cli)
+            asyncio.run(p.chat([{"role": "user", "content": "hi"}], "sys"))
+            _, kwargs = cli.chat.completions.create.call_args
+            assert kwargs.get("max_tokens") == 4096