From 3133a3b32fb6aa491c74b43c66169dd7c587b529 Mon Sep 17 00:00:00 2001 From: Sirius Claw Date: Wed, 22 Apr 2026 06:32:35 +0000 Subject: [PATCH 1/6] fix(vllm): update to recommended qwen3_coder tool parser, auto max-model-len, and latest image Addresses tokenizer backend / parser mismatch on heretic-style Qwen3.5 models per model card recommendations. --- apps/base/customer1/openclaw/a100-vllm.yaml | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/apps/base/customer1/openclaw/a100-vllm.yaml b/apps/base/customer1/openclaw/a100-vllm.yaml index 7fe4336..e6fbcf4 100644 --- a/apps/base/customer1/openclaw/a100-vllm.yaml +++ b/apps/base/customer1/openclaw/a100-vllm.yaml @@ -24,7 +24,7 @@ spec: effect: "NoSchedule" containers: - name: vllm-brain - image: vllm/vllm-openai:v0.19.1-ubuntu2404 + image: vllm/vllm-openai:latest command: ["python3", "-m", "vllm.entrypoints.openai.api_server"] env: - name: HF_TOKEN @@ -36,12 +36,12 @@ spec: - --tensor-parallel-size=1 - --gpu-memory-utilization=0.95 - --kv-cache-dtype=fp8 - - --max-model-len=49152 + - --max-model-len=auto - --enable-auto-tool-choice - --trust-remote-code - --dtype=auto - --enable-prefix-caching - - --tool-call-parser=hermes + - --tool-call-parser=qwen3_coder - --reasoning-parser=qwen3 - --disable-custom-all-reduce From 41f0b43d3b6c3d941f3bfd78127b5d138db80a03 Mon Sep 17 00:00:00 2001 From: Sirius Claw Date: Wed, 22 Apr 2026 12:45:10 +0000 Subject: [PATCH 2/6] optimize: trim SOUL.md by ~65%, reduce context injection to curated files only, increase requestTimeout to 180s, enable promptCaching per optimization guide - SOUL.md now <600 bytes (core invariants only) - Added context.injectAllMarkdown=false + explicit curated list - requestTimeoutSeconds=180 for high-param models - promptCaching=true to avoid reprocessing system prompt every turn - Fixed missing comma in agents.defaults.model (was invalid JSON) Per https://github.com/OnlyTerp/openclaw-optimization-guide --- apps/base/customer1/openclaw/configmap.yaml | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/apps/base/customer1/openclaw/configmap.yaml b/apps/base/customer1/openclaw/configmap.yaml index 56e0f7f..b8dc36f 100644 --- a/apps/base/customer1/openclaw/configmap.yaml +++ b/apps/base/customer1/openclaw/configmap.yaml @@ -62,7 +62,7 @@ data: "agents": { "defaults": { "model": { - "primary": "xai/grok-4.1-fast" + "primary": "xai/grok-4.1-fast", "fallbacks": ["xai/grok-4.1-fast"] }, "workspace": "/home/node/.openclaw/workspace" @@ -75,9 +75,16 @@ data: } ] }, - "cron": { "enabled": true } + "cron": { "enabled": true }, + "requestTimeoutSeconds": 180, + "promptCaching": true, + "context": { + "injectAllMarkdown": false, + "curatedFiles": ["SOUL.md", "MEMORY.md", "IDENTITY.md", "USER.md"], + "useMemorySearch": true + } } AGENTS.md: | - # OpenClaw Assistant + # Sirius Claw - You are a helpful AI assistant running in Kubernetes. + Professional, precise, dark humor. Running in Lance's GKE cluster. From 240620c3a4728d1faa04cd1e2d449884e39bcf19 Mon Sep 17 00:00:00 2001 From: Sirius Claw Date: Wed, 22 Apr 2026 17:07:05 +0000 Subject: [PATCH 3/6] feat: add KEDA scaling for A100 vLLM - Scale to 0 when idle (saves money) - 15 min cooldown before scale-down - Triggers on 3+ pending HTTP requests - Targets openclaw-brain-vllm deployment --- apps/base/customer1/openclaw/keda-vllm.yaml | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) create mode 100644 apps/base/customer1/openclaw/keda-vllm.yaml diff --git a/apps/base/customer1/openclaw/keda-vllm.yaml b/apps/base/customer1/openclaw/keda-vllm.yaml new file mode 100644 index 0000000..09dadc2 --- /dev/null +++ b/apps/base/customer1/openclaw/keda-vllm.yaml @@ -0,0 +1,21 @@ +apiVersion: keda.sh/v1alpha1 +kind: ScaledObject +metadata: + name: openclaw-brain-a100-scaling + namespace: customer1 +spec: + scaleTargetRef: + name: openclaw-brain-vllm + kind: Deployment + service: + name: openclaw-brain-service + port: 8000 + minReplicaCount: 0 + maxReplicaCount: 1 + cooldownPeriod: 900 # 15 minutes idle before scale-down + triggers: + - type: http + metadata: + path: "/v1/models" + host: "openclaw-brain-service.customer1.svc.cluster.local" + requestCount: "3" # Scale up when 3+ requests pending From 9673b76fc74f04281235a56d098b1cfd1d5ec2c4 Mon Sep 17 00:00:00 2001 From: Sirius Claw Date: Wed, 22 Apr 2026 17:22:22 +0000 Subject: [PATCH 4/6] fix: trigger KEDA scale-up on 1 request instead of 3 - Wake script handles cold start, so scale immediately on first request - Reduces latency for /wake command and manual requests --- apps/base/customer1/openclaw/keda-vllm.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/apps/base/customer1/openclaw/keda-vllm.yaml b/apps/base/customer1/openclaw/keda-vllm.yaml index 09dadc2..c26cf69 100644 --- a/apps/base/customer1/openclaw/keda-vllm.yaml +++ b/apps/base/customer1/openclaw/keda-vllm.yaml @@ -18,4 +18,4 @@ spec: metadata: path: "/v1/models" host: "openclaw-brain-service.customer1.svc.cluster.local" - requestCount: "3" # Scale up when 3+ requests pending + requestCount: "1" # Scale up on 1+ request (wake script handles cold start) From 8b0e1ec33245819ee30276cb8e29e62f764d4276 Mon Sep 17 00:00:00 2001 From: sirius0xdev Date: Wed, 22 Apr 2026 13:35:12 -0400 Subject: [PATCH 5/6] Update kustomization.yaml --- apps/base/customer1/openclaw/kustomization.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/apps/base/customer1/openclaw/kustomization.yaml b/apps/base/customer1/openclaw/kustomization.yaml index 546fdd1..e278e66 100644 --- a/apps/base/customer1/openclaw/kustomization.yaml +++ b/apps/base/customer1/openclaw/kustomization.yaml @@ -2,7 +2,7 @@ apiVersion: kustomize.config.k8s.io/v1beta1 kind: Kustomization resources: # - vllm-l4.yaml - #- keda-a100.yaml + - keda-a100.yaml - pvc.yaml - configmap.yaml - deployment.yaml From 283f3eba26be21e7b7226d3811e3436d4cb417d1 Mon Sep 17 00:00:00 2001 From: sirius0xdev Date: Wed, 22 Apr 2026 14:47:40 -0400 Subject: [PATCH 6/6] Update kustomization.yaml --- apps/base/customer1/openclaw/kustomization.yaml | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/apps/base/customer1/openclaw/kustomization.yaml b/apps/base/customer1/openclaw/kustomization.yaml index e278e66..5c7138d 100644 --- a/apps/base/customer1/openclaw/kustomization.yaml +++ b/apps/base/customer1/openclaw/kustomization.yaml @@ -2,7 +2,7 @@ apiVersion: kustomize.config.k8s.io/v1beta1 kind: Kustomization resources: # - vllm-l4.yaml - - keda-a100.yaml + - keda-vllm.yaml - pvc.yaml - configmap.yaml - deployment.yaml