From c4ea5e3a9858b2109149b871c0374b29cbf9b001 Mon Sep 17 00:00:00 2001
From: Abiba <abiba@sysloggh.com>
Date: Wed, 27 May 2026 00:01:33 +0000
Subject: [PATCH] fix: flip Tier 4 (Heavy) to Dense-first for thermal safety
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Dense → MoE → VLM instead of MoE → Dense → VLM.
Combined with MoE at 1 concurrent slot, Dense absorbs all
primary traffic. MoE only activates when Dense saturated.
Prevents Strix Halo from hitting 94C thermal limit.
---
 router/router.py | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/router/router.py b/router/router.py
index 3a02642..2423133 100644
--- a/router/router.py
+++ b/router/router.py
@@ -233,9 +233,9 @@ def route(rd, tier):
         result = select_best_gpu(candidates, "medium")
         if result: return result
     
-    # TIER 4: Heavy reasoning — large context or very long conversations → MoE first
+    # TIER 4: Heavy reasoning — Dense first (thermal), MoE fallback (192K/262K context)
     if t > 25000 or turns > 15:
-        candidates = [m for m in ["qwen3.6-35B-A3B","qwen3.6-27B-code","qwen3.5-9b-vlm"] if m in avail]
+        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