Skip to content

Change vLLM app config for Qwwen 3.8 27B #3

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
nwthomas merged 1 commit into from
Aug 21, 2026
Merged

Change vLLM app config for Qwwen 3.8 27B #3

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
30 changes: 20 additions & 10 deletions helm/vllm/values.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5,18 +5,28 @@ replicaCount: 1

image:
repository: vllm/vllm-openai
tag: "v0.12.0"
tag: "v0.27.1"
pullPolicy: Always

vllm:
port: 8000
cacheMountPath: /root/.cache/huggingface
shmSizeLimit: "2Gi"
shmSizeLimit: "16Gi"
command:
- /bin/sh
- -c
args:
- "vllm serve openai/gpt-oss-20b --trust-remote-code --enable-chunked-prefill --enable-auto-tool-choice --tool-call-parser openai --reasoning-parser openai_gptoss"
- >-
vllm serve unsloth/Qwen3.8-27B-NVFP4
--served-model-name qwen3.8-27b
--tensor-parallel-size 1
--max-model-len 32768
--kv-cache-dtype fp8
--gpu-memory-utilization 0.90
--enforce-eager
--reasoning-parser qwen3
--enable-auto-tool-choice
--tool-call-parser qwen3_coder
gpu:
enabled: true
type: "nvidia"
Expand Down Expand Up @@ -82,17 +92,17 @@ ingress:
resources:
requests:
nvidia.com/gpu: "1"
memory: "6G"
cpu: "2"
memory: "32G"
cpu: "4"
limits:
nvidia.com/gpu: "1"
memory: "20G"
cpu: "10"
memory: "96G"
cpu: "12"

livenessProbe:
enabled: true
path: /health
initialDelaySeconds: 60
initialDelaySeconds: 300

@chatgpt-codex-connector chatgpt-codex-connector Bot Aug 21, 2026

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Badge Protect cold model loading with a startup probe

When the 27B model is not already cached or initialization takes more than roughly 320 seconds, the liveness probe configured in helm/vllm/templates/deployment.yaml begins after this delay and restarts the container after three 10-second failures. The increased readiness failure threshold cannot prevent that restart, so a cold deployment can repeatedly interrupt the model download/load; use a startup probe or give liveness a cold-start window long enough for this model.

Useful? React with 👍 / 👎.

periodSeconds: 10
timeoutSeconds: 10
failureThreshold: 3
Expand All @@ -101,10 +111,10 @@ livenessProbe:
readinessProbe:
enabled: true
path: /health
initialDelaySeconds: 60
initialDelaySeconds: 300
periodSeconds: 5
timeoutSeconds: 5
failureThreshold: 3
failureThreshold: 24
successThreshold: 1

autoscaling:
Expand Down
Loading