From 32bd817e97fc6fa89af90eb0f9e6d4a97ca17da5 Mon Sep 17 00:00:00 2001 From: Abiba Date: Tue, 19 May 2026 21:24:36 +0000 Subject: [PATCH] =?UTF-8?q?fix:=20heavy=20tier=20back=20to=20Dense?= =?UTF-8?q?=E2=86=92MoE=E2=86=92VLM=20(Dense=20now=2098K)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit --- router/router.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/router/router.py b/router/router.py index 80b9f69..bfd8d7a 100644 --- a/router/router.py +++ b/router/router.py @@ -190,8 +190,8 @@ def route(rd, tier): # TIER 3: Heavy reasoning — extremely large context or very long conversations if t > 50000 or turns > 25: - # Prefer models with larger context windows (MoE/VLM at 131K, Dense at 65K) - candidates = [m for m in ["qwen3.6-35B-A3B","qwen3.5-9b-vlm","qwen3.6-27B-code"] if m in avail] + # Dense first (98K, purpose-built for reasoning), then MoE/VLM 131K + candidates = [m for m in ["qwen3.6-27B-code","qwen3.6-35B-A3B","qwen3.5-9b-vlm"] if m in avail] result = select_best_gpu(candidates, "heavy_reasoning") if result: return result