diff --git a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml index ab2388b..1a7e92f 100644 --- a/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml +++ b/infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml @@ -39,11 +39,11 @@ spec: - --tensor-parallel-size=1 - --tokenizer-mode=hf - --gpu-memory-utilization=0.95 - - --max-model-len=262144 + - --max-model-len=131072 - --enable-auto-tool-choice - --enable-chunked-prefill - - --max-num-batched-tokens=8192 - - --max-num-seqs=64 + - --max-num-batched-tokens=4096 + - --max-num-seqs=8 - --trust-remote-code - --dtype=auto - --enable-prefix-caching