From cfedb963e532f1259e344e51f86b1b97d94d7e74 Mon Sep 17 00:00:00 2001
From: sirius0xdev <sirius0xdev@proton.me>
Date: Sun, 10 May 2026 03:02:51 +0000
Subject: [PATCH] dial in vllm

---
 .../gpus/base/vllm-servers/rtx6000-vllm.yaml           | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
index 32947b5..c490324 100644
--- a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
+++ b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
@@ -37,13 +37,13 @@ spec:
             - --port=8000
             - --tensor-parallel-size=1
             - --tokenizer-mode=hf 
-            - --gpu-memory-utilization=0.95 
-            - --enforce-eager 
-            - --max-model-len=65536
+            - --gpu-memory-utilization=0.90  
+            #- --enforce-eager 
+            - --max-model-len=98304 
             - --enable-auto-tool-choice 
             - --kv-cache-dtype=fp8 
-            - --max-num-batched-tokens=65536 
-            - --max-num-seqs=128
+            - --max-num-batched-tokens=32768 
+            - --max-num-seqs=32
             - --block-size=32 
             - --enable-chunked-prefill
             - --trust-remote-code