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

Merge pull request #27 from sirius0xdev/optimize-vllm-a100

Browse source 
chore: Optimize vLLM for A100 by removing artificial batch limits
This commit is contained in:
sirius0xdev 2026-04-19 22:29:24 -04:00 • committed by GitHub
commit ea8ea2801d
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
1 changed files with 0 additions and 3 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
apps/base/customer1/openclaw/vllm-gemma.yaml
View file

@ -38,9 +38,6 @@ spec:
- --tensor-parallel-size=1 - --tensor-parallel-size=1
- --gpu-memory-utilization=0.95 # ← Critical: lower than default 0.90 - --gpu-memory-utilization=0.95 # ← Critical: lower than default 0.90
- --max-model-len=24576 # Start conservative (you can raise to 32768 later) - --max-model-len=24576 # Start conservative (you can raise to 32768 later)
- --max-num-batched-tokens=8192
- --max-num-seqs=8 # Lower for single-user / low-concurrency
- --disable-custom-all-reduce
- --enable-auto-tool-choice - --enable-auto-tool-choice
- --trust-remote-code - --trust-remote-code
- --dtype=auto - --dtype=auto