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/config.py
+++ b/api/config.py
@@ -20,7 +20,8 @@ class Settings(BaseSettings):
    LLM_API_KEY: str = os.getenv("LLM_API_KEY", "")
    LLM_MODEL: str = os.getenv("LLM_MODEL", "")
    LLM_TIMEOUT: int = int(os.getenv("LLM_TIMEOUT", "120"))
-    LLM_MAX_TOKENS: int = int(os.getenv("LLM_MAX_TOKENS", "1000"))
+    # reasoning-модели тратят max_tokens и на размышления — 1000 им мало
+    LLM_MAX_TOKENS: int = int(os.getenv("LLM_MAX_TOKENS", "8000"))
    LLM_TEMPERATURE: float = float(os.getenv("LLM_TEMPERATURE", "0.3"))

    @property