Merge pull request #38 from sirius0xdev/feat/optimize-vllm-cost-perf

feat: optimize cost via L4 spot instances and vLLM performance tuning
This commit is contained in:
sirius0xdev 2026-04-20 01:39:35 -04:00 committed by GitHub
commit 70ad49f21f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
3 changed files with 5 additions and 3 deletions

View file

@ -24,7 +24,7 @@ spec:
effect: "NoSchedule" effect: "NoSchedule"
containers: containers:
- name: vllm-brain - name: vllm-brain
image: vllm/vllm-openai:latest image: vllm/vllm-openai:v0.4.2
command: ["python3", "-m", "vllm.entrypoints.openai.api_server"] command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
env: env:
- name: HF_TOKEN - name: HF_TOKEN
@ -44,6 +44,8 @@ spec:
- --tool-call-parser=qwen3_coder - --tool-call-parser=qwen3_coder
- --reasoning-parser=qwen3 - --reasoning-parser=qwen3
- --enable-prefix-caching - --enable-prefix-caching
- --enable-chunked-prefill
- --kv-cache-dtype=fp8
##- --limit-mm-per-prompt='{"image"=1, "video"=0}' # Optional: limit vision if not heavily using images yet ##- --limit-mm-per-prompt='{"image"=1, "video"=0}' # Optional: limit vision if not heavily using images yet
ports: ports:
- containerPort: 8000 - containerPort: 8000

View file

@ -24,7 +24,7 @@ spec:
effect: "NoSchedule" effect: "NoSchedule"
containers: containers:
- name: vllm-brain-l4 - name: vllm-brain-l4
image: vllm/vllm-openai:latest image: vllm/vllm-openai:v0.4.2
command: ["python3", "-m", "vllm.entrypoints.openai.api_server"] command: ["python3", "-m", "vllm.entrypoints.openai.api_server"]
env: env:
- name: HF_TOKEN - name: HF_TOKEN

View file

@ -23,7 +23,7 @@ resource "google_container_node_pool" "gpu_pool" {
} }
spot = false spot = true
oauth_scopes = [ oauth_scopes = [
"https://www.googleapis.com/auth/cloud-platform" "https://www.googleapis.com/auth/cloud-platform"