From 93d0d3cc4b158be9e5667bd14c63e1e37573c2ef Mon Sep 17 00:00:00 2001 From: Abiba Date: Wed, 27 May 2026 00:04:42 +0000 Subject: [PATCH] revert: MoE concurrency back to 2 (Dense-first routing handles thermal) --- router/router.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/router/router.py b/router/router.py index 2423133..5af41d8 100644 --- a/router/router.py +++ b/router/router.py @@ -19,7 +19,7 @@ GPU_URLS = { } # Max concurrent requests per GPU (based on llama.cpp --parallel) GPU_MAX_CONCURRENT = { - "qwen3.6-35B-A3B": 1, # 1 slot (thermal management: 94C at 2 concurrent) + "qwen3.6-35B-A3B": 2, # 2 slots (Dense-first routing reduces thermal load) "qwen3.6-27B-code": 1, # 1 slot (24GB VRAM saturated at 256K ctx) "qwen3.5-9b-vlm": 2, # 2 slots (12GB VRAM, 4GB headroom) }