From c580c935acd10b026e01d631fca24b262d3dde08 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?=E2=80=9CNaeel=E2=80=9D?= <tazet@narod.ru>
Date: Sun, 14 Jun 2026 07:22:10 +0400
Subject: [PATCH] =?UTF-8?q?fix(llm):=20=D0=BC=D0=BE=D0=B4=D0=B5=D0=BB?=
 =?UTF-8?q?=D1=8C=20gpt-oss-120b,=20max=5Ftokens=208000?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Доступные модели на api.aillm.ru: whisper, qwen3.6-27b, gpt-oss-120b.
gpt-oss-120b — та самая 120B модель.
max_tokens поднят до 8000 (reasoning съедает часть токенов).
---
 site/llm_client.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/site/llm_client.py b/site/llm_client.py
index 7d299af..9db5a7a 100644
--- a/site/llm_client.py
+++ b/site/llm_client.py
@@ -15,10 +15,10 @@ import requests
 # ── Конфигурация ────────────────────────────────────────────────
 
 DEFAULT_URL = "https://api.aillm.ru/v1/chat/completions"
-DEFAULT_MODEL = "deepseek-chat"  # DeepSeek V3 на aillm.ru (120B эквивалент)
+DEFAULT_MODEL = "gpt-oss-120b"  # 120B модель через LiteLLM-прокси
 
 
-def ask(prompt: str, model: str = None, max_tokens: int = 4000) -> dict:
+def ask(prompt: str, model: str = None, max_tokens: int = 8000) -> dict:
     """
     Отправить промпт к LLM API.