sirius/gcloud-lab
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

change vllm settings

Browse source
This commit is contained in:
sirius0xdev 2026-05-10 01:33:47 +00:00
parent d98d03981e
commit 3d3f0a5a1f
2 changed files with 6 additions and 5 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

2
apps/base/customer1/hermes-agent/configmap.yaml
View file

@ -14,7 +14,7 @@ data:
type: openai type: openai
api_key: EMPTY api_key: EMPTY
base_url: http://rtx6000-brain-service.customer1.svc.cluster.local:8000/v1 base_url: http://rtx6000-brain-service.customer1.svc.cluster.local:8000/v1
context_length: 131072 context_length: 176072
qwen-vllm: qwen-vllm:
type: openai type: openai
api_key: EMPTY api_key: EMPTY

9
infrastructure/gpus/base/vllm-servers/rtx6000-vllm.yaml
View file

@ -38,11 +38,12 @@ spec:
- --tensor-parallel-size=1 - --tensor-parallel-size=1
- --tokenizer-mode=hf - --tokenizer-mode=hf
- --gpu-memory-utilization=0.95 - --gpu-memory-utilization=0.95
- --max-model-len=237864 - --max-model-len=98304
- --enable-auto-tool-choice - --enable-auto-tool-choice
- --kv-cache-dtype=fp8 - --kv-cache-dtype=fp8
- --max-num-batched-tokens=32768 - --max-num-batched-tokens=65536
- --max-num-seqs=16 - --max-num-seqs=128
- --enable-chunked-prefill
- --trust-remote-code - --trust-remote-code
- --dtype=auto - --dtype=auto
- --enable-prefix-caching - --enable-prefix-caching
@ -59,7 +60,7 @@ spec:
requests: requests:
nvidia.com/gpu: 1 nvidia.com/gpu: 1
memory: "150Gi" memory: "150Gi"
cpu: "24" cpu: "32"
startupProbe: startupProbe:
httpGet: httpGet: