dial in vllm

2026-05-10 03:02:51 +00:00 · 2026-05-10 03:02:51 +00:00 · cfedb963e5
commit cfedb963e5
parent 4e2b0974f4
1 changed files with 5 additions and 5 deletions
--- 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