LLM max_tokens 1000→8000: reasoning-модели тратили весь лимит на размышления и возвращали пустой ответ

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
2026-07-14 15:39:47 +05:00
parent c88b0e4f6d
commit 31eb8c3303
4 changed files with 6 additions and 4 deletions
--- a/api/lib/llm.py
+++ b/api/lib/llm.py
@@ -36,7 +36,8 @@ class LLMConfig(BaseModel):
    # before the user has picked one.
    model: str = ""
    timeout: int = 120
-    max_tokens: int = 1000
+    # у reasoning-моделей сюда входят и токены размышлений
+    max_tokens: int = 8000
    temperature: float = 0.3
    extra_headers: dict[str, str] = Field(default_factory=dict)