From 621a897bec279732448dbe15c78eb31e80c994f2 Mon Sep 17 00:00:00 2001 From: Abiba Date: Wed, 27 May 2026 00:29:25 +0000 Subject: [PATCH] =?UTF-8?q?tune:=20raise=20Tier=202=20threshold=204K?= =?UTF-8?q?=E2=86=9210K=20tok,=206=E2=86=9210=20turns=20for=20VLM?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit More conversations now route to VLM as primary. 9B VLM has 262K context window and 88 tok/s average — well suited for moderate conversations. Dense absorbs overflow and heavy reasoning. --- router/router.py | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/router/router.py b/router/router.py index 5af41d8..de0d340 100644 --- a/router/router.py +++ b/router/router.py @@ -218,8 +218,8 @@ def route(rd, tier): result = select_best_gpu(fallback, "lightweight_fallback") if result: return result - # TIER 2: Simple conversations — short context, any prompt → VLM first, Dense second - if t <= 4000 and turns <= 6 and "qwen3.5-9b-vlm" in avail: + # TIER 2: Simple conversations — VLM primary (up to 10K tok), any prompt → VLM first, Dense second + if t <= 10000 and turns <= 10 and "qwen3.5-9b-vlm" in avail: if not is_gpu_busy("qwen3.5-9b-vlm"): return {"model":"qwen3.5-9b-vlm","reason":"simple_conv"} # VLM busy — fall back to Dense, then MoE