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/route/pipeline.py
+++ b/api/route/pipeline.py
@@ -18,7 +18,7 @@ class LLMOverride(BaseModel):
    api_key: str = ""
    model: str
    timeout: int = 120
-    max_tokens: int = 1000
+    max_tokens: int = 8000
    temperature: float = 0.3
    extra_headers: dict[str, str] = {}
    system_prompt: str | None = None