[PATCH v2 4/4] defconfigs: Add composable fragments for Lambda Labs vLLM deployment

This introduces a fragment-based approach to defconfig composition, allowing
users to combine infrastructure provisioning with workflow configurations.

Two new config fragments are added to defconfigs/configs/:

- lambdalabs-gpu-1x-a10.config: Terraform configuration for Lambda Labs A10
  GPU instance provisioning with automatic region inference and SSH key
  generation.

- vllm-production-stack-gpu.config: vLLM production stack configuration with
  GPU-accelerated inference, Kubernetes deployment via minikube, monitoring,
  autoscaling, and benchmarking capabilities.

These fragments are combined into a new defconfig lambdalabs-vllm-gpu-1x-a10
which enables end-to-end deployment: provision a Lambda Labs A10 GPU instance
($0.75/hr) and deploy the vLLM production stack for LLM inference workloads.

The fragment approach allows users to compose configurations by combining
infrastructure providers (Lambda Labs, AWS, Azure, bare metal) with different
workflows (vLLM, fstests, blktests) without maintaining separate defconfigs
for every combination.

Example usage:
  make defconfig-lambdalabs-vllm-gpu-1x-a10
  make bringup        # Provisions Lambda Labs A10 GPU instance
  make vllm           # Deploys vLLM production stack
  make vllm-benchmark # Run performance benchmarks

Generated-by: Claude AI
Signed-off-by: Luis Chamberlain <[email protected]>
---
 .../configs/lambdalabs-gpu-1x-a10.config      |   8 ++
 .../configs/vllm-production-stack-gpu.config  |  61 +++++++++++
 defconfigs/lambdalabs-vllm-gpu-1x-a10         | 103 ++++++++++++++++++
 3 files changed, 172 insertions(+)
 create mode 100644 defconfigs/configs/lambdalabs-gpu-1x-a10.config
 create mode 100644 defconfigs/configs/vllm-production-stack-gpu.config
 create mode 100644 defconfigs/lambdalabs-vllm-gpu-1x-a10

diff --git a/defconfigs/configs/lambdalabs-gpu-1x-a10.config b/defconfigs/configs/lambdalabs-gpu-1x-a10.config
new file mode 100644
index 00000000..c85dae4e
--- /dev/null
+++ b/defconfigs/configs/lambdalabs-gpu-1x-a10.config
@@ -0,0 +1,8 @@
+# Lambda Labs GPU 1x A10 instance configuration
+CONFIG_TERRAFORM=y
+CONFIG_TERRAFORM_LAMBDALABS=y
+CONFIG_TERRAFORM_LAMBDALABS_REGION_SMART_INFER=y
+CONFIG_TERRAFORM_LAMBDALABS_INSTANCE_TYPE_GPU_1X_A10=y
+CONFIG_TERRAFORM_SSH_CONFIG_GENKEY=y
+CONFIG_TERRAFORM_SSH_CONFIG_GENKEY_OVERWRITE=y
+CONFIG_TERRAFORM_SSH_CONFIG_GENKEY_EMPTY_PASSPHRASE=y
diff --git a/defconfigs/configs/vllm-production-stack-gpu.config b/defconfigs/configs/vllm-production-stack-gpu.config
new file mode 100644
index 00000000..75b11a9f
--- /dev/null
+++ b/defconfigs/configs/vllm-production-stack-gpu.config
@@ -0,0 +1,61 @@
+# vLLM Production Stack with GPU support
+CONFIG_WORKFLOWS=y
+CONFIG_WORKFLOWS_TESTS=y
+CONFIG_WORKFLOWS_LINUX_TESTS=y
+CONFIG_WORKFLOWS_DEDICATED_WORKFLOW=y
+CONFIG_KDEVOPS_WORKFLOW_DEDICATE_VLLM=y
+
+# vLLM Production Stack with Kubernetes
+CONFIG_VLLM_PRODUCTION_STACK=y
+CONFIG_VLLM_K8S_MINIKUBE=y
+CONFIG_VLLM_VERSION_STABLE=y
+CONFIG_VLLM_ENGINE_IMAGE_TAG="v0.10.2"
+CONFIG_VLLM_HELM_RELEASE_NAME="vllm-prod"
+CONFIG_VLLM_HELM_NAMESPACE="vllm-system"
+
+# Production Stack components
+CONFIG_VLLM_PROD_STACK_REPO="https://vllm-project.github.io/production-stack"
+CONFIG_VLLM_PROD_STACK_CHART_VERSION="latest"
+CONFIG_VLLM_PROD_STACK_ROUTER_IMAGE="ghcr.io/vllm-project/production-stack/router"
+CONFIG_VLLM_PROD_STACK_ROUTER_TAG="latest"
+CONFIG_VLLM_PROD_STACK_ENABLE_MONITORING=y
+CONFIG_VLLM_PROD_STACK_ENABLE_AUTOSCALING=y
+CONFIG_VLLM_PROD_STACK_MIN_REPLICAS=2
+CONFIG_VLLM_PROD_STACK_MAX_REPLICAS=5
+CONFIG_VLLM_PROD_STACK_TARGET_GPU_UTILIZATION=80
+
+# Model configuration
+CONFIG_VLLM_MODEL_URL="facebook/opt-125m"
+CONFIG_VLLM_MODEL_NAME="opt-125m"
+
+# GPU configuration - EXPLICITLY DISABLED CPU INFERENCE
+# CONFIG_VLLM_USE_CPU_INFERENCE is not set
+CONFIG_VLLM_REQUEST_GPU=1
+CONFIG_VLLM_GPU_TYPE=""
+CONFIG_VLLM_GPU_MEMORY_UTILIZATION="0.5"
+CONFIG_VLLM_TENSOR_PARALLEL_SIZE=1
+
+# Engine configuration for GPU
+CONFIG_VLLM_REPLICA_COUNT=1
+CONFIG_VLLM_REQUEST_CPU=8
+CONFIG_VLLM_REQUEST_MEMORY="16Gi"
+CONFIG_VLLM_MAX_MODEL_LEN=1024
+CONFIG_VLLM_DTYPE="auto"
+
+# Router and observability
+CONFIG_VLLM_ROUTER_ENABLED=y
+CONFIG_VLLM_ROUTER_ROUND_ROBIN=y
+CONFIG_VLLM_OBSERVABILITY_ENABLED=y
+CONFIG_VLLM_GRAFANA_PORT=3000
+CONFIG_VLLM_PROMETHEUS_PORT=9090
+
+# API configuration
+CONFIG_VLLM_API_PORT=8000
+CONFIG_VLLM_API_KEY=""
+CONFIG_VLLM_HF_TOKEN=""
+
+# Benchmarking
+CONFIG_VLLM_BENCHMARK_ENABLED=y
+CONFIG_VLLM_BENCHMARK_DURATION=60
+CONFIG_VLLM_BENCHMARK_CONCURRENT_USERS=10
+CONFIG_VLLM_BENCHMARK_RESULTS_DIR="/data/vllm-benchmark"
diff --git a/defconfigs/lambdalabs-vllm-gpu-1x-a10 b/defconfigs/lambdalabs-vllm-gpu-1x-a10
new file mode 100644
index 00000000..926be1bd
--- /dev/null
+++ b/defconfigs/lambdalabs-vllm-gpu-1x-a10
@@ -0,0 +1,103 @@
+#
+# Lambda Labs vLLM Production Stack - 1x A10 GPU ($0.75/hr)
+#
+# This combines:
+#   - defconfigs/configs/lambdalabs-gpu-1x-a10.config (Terraform provisioning)
+#   - defconfigs/configs/vllm-production-stack-gpu.config (vLLM deployment)
+#
+# Provisions a Lambda Labs GPU instance with NVIDIA A10 (24GB) and deploys
+# the vLLM production stack for LLM inference workloads.
+#
+# ============================================================================
+# NVIDIA GPU COMPATIBILITY (CUDA):
+# ============================================================================
+#
+# vLLM v0.10.x uses FlashInfer CUDA kernels that require NVIDIA GPUs with
+# compute capability >= 8.0. Older NVIDIA GPUs will fail with:
+#   "RuntimeError: TopPSamplingFromProbs failed with error code
+#    too many resources requested for launch"
+#
+# NVIDIA A10 Compatibility:
+#   - Compute Capability: 8.6 ✓ COMPATIBLE
+#   - Memory: 24GB GDDR6
+#   - Cost: $0.75/hour on Lambda Labs
+#   - Perfect for: Production LLM inference, fine-tuning
+#
+# ============================================================================
+# Usage:
+#   make defconfig-lambdalabs-vllm-gpu-1x-a10
+#   make bringup        # Provisions A10 GPU instance
+#   make vllm           # Deploys vLLM production stack
+#   make vllm-benchmark # Run performance benchmarks
+# ============================================================================
+#
+# Lambda Labs GPU 1x A10 instance configuration
+CONFIG_TERRAFORM=y
+CONFIG_TERRAFORM_LAMBDALABS=y
+CONFIG_TERRAFORM_LAMBDALABS_REGION_SMART_INFER=y
+CONFIG_TERRAFORM_LAMBDALABS_INSTANCE_TYPE_GPU_1X_A10=y
+CONFIG_TERRAFORM_SSH_CONFIG_GENKEY=y
+CONFIG_TERRAFORM_SSH_CONFIG_GENKEY_OVERWRITE=y
+CONFIG_TERRAFORM_SSH_CONFIG_GENKEY_EMPTY_PASSPHRASE=y
+
+# vLLM Production Stack with GPU support
+CONFIG_WORKFLOWS=y
+CONFIG_WORKFLOWS_TESTS=y
+CONFIG_WORKFLOWS_LINUX_TESTS=y
+CONFIG_WORKFLOWS_DEDICATED_WORKFLOW=y
+CONFIG_KDEVOPS_WORKFLOW_DEDICATE_VLLM=y
+
+# vLLM Production Stack with Kubernetes
+CONFIG_VLLM_PRODUCTION_STACK=y
+CONFIG_VLLM_K8S_MINIKUBE=y
+CONFIG_VLLM_VERSION_STABLE=y
+CONFIG_VLLM_ENGINE_IMAGE_TAG="v0.10.2"
+CONFIG_VLLM_HELM_RELEASE_NAME="vllm-prod"
+CONFIG_VLLM_HELM_NAMESPACE="vllm-system"
+
+# Production Stack components
+CONFIG_VLLM_PROD_STACK_REPO="https://vllm-project.github.io/production-stack"
+CONFIG_VLLM_PROD_STACK_CHART_VERSION="latest"
+CONFIG_VLLM_PROD_STACK_ROUTER_IMAGE="ghcr.io/vllm-project/production-stack/router"
+CONFIG_VLLM_PROD_STACK_ROUTER_TAG="latest"
+CONFIG_VLLM_PROD_STACK_ENABLE_MONITORING=y
+CONFIG_VLLM_PROD_STACK_ENABLE_AUTOSCALING=y
+CONFIG_VLLM_PROD_STACK_MIN_REPLICAS=2
+CONFIG_VLLM_PROD_STACK_MAX_REPLICAS=5
+CONFIG_VLLM_PROD_STACK_TARGET_GPU_UTILIZATION=80
+
+# Model configuration
+CONFIG_VLLM_MODEL_URL="facebook/opt-125m"
+CONFIG_VLLM_MODEL_NAME="opt-125m"
+
+# GPU configuration - EXPLICITLY DISABLED CPU INFERENCE
+# CONFIG_VLLM_USE_CPU_INFERENCE is not set
+CONFIG_VLLM_REQUEST_GPU=1
+CONFIG_VLLM_GPU_TYPE=""
+CONFIG_VLLM_GPU_MEMORY_UTILIZATION="0.5"
+CONFIG_VLLM_TENSOR_PARALLEL_SIZE=1
+
+# Engine configuration for GPU
+CONFIG_VLLM_REPLICA_COUNT=1
+CONFIG_VLLM_REQUEST_CPU=8
+CONFIG_VLLM_REQUEST_MEMORY="16Gi"
+CONFIG_VLLM_MAX_MODEL_LEN=1024
+CONFIG_VLLM_DTYPE="auto"
+
+# Router and observability
+CONFIG_VLLM_ROUTER_ENABLED=y
+CONFIG_VLLM_ROUTER_ROUND_ROBIN=y
+CONFIG_VLLM_OBSERVABILITY_ENABLED=y
+CONFIG_VLLM_GRAFANA_PORT=3000
+CONFIG_VLLM_PROMETHEUS_PORT=9090
+
+# API configuration
+CONFIG_VLLM_API_PORT=8000
+CONFIG_VLLM_API_KEY=""
+CONFIG_VLLM_HF_TOKEN=""
+
+# Benchmarking
+CONFIG_VLLM_BENCHMARK_ENABLED=y
+CONFIG_VLLM_BENCHMARK_DURATION=60
+CONFIG_VLLM_BENCHMARK_CONCURRENT_USERS=10
+CONFIG_VLLM_BENCHMARK_RESULTS_DIR="/data/vllm-benchmark"
-- 
2.51.0