From ad30b9524f8ba57394d659db0615d20c95ebdb37 Mon Sep 17 00:00:00 2001
From: sirius0xdev <sirius0xdev@proton.me>
Date: Sun, 10 May 2026 02:20:42 +0000
Subject: [PATCH] numseq lower

---
 infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
index c3d58d7..6e105d6 100644
--- a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
+++ b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
@@ -37,12 +37,12 @@ spec:
             - --port=8000
             - --tensor-parallel-size=1
             - --tokenizer-mode=hf 
-            - --gpu-memory-utilization=0.90 
+            - --gpu-memory-utilization=0.95 
             - --max-model-len=65536
             - --enable-auto-tool-choice 
             - --kv-cache-dtype=fp8 
             - --max-num-batched-tokens=65536 
-            - --max-num-seqs=64 
+            - --max-num-seqs=32  
             - --enable-chunked-prefill
             - --trust-remote-code
             - --dtype=auto