51 lines
1.9 KiB
YAML
51 lines
1.9 KiB
YAML
# Static, outbound-only M4 runtime worker slot for a remote GPU node.
|
|
services:
|
|
runtime-worker:
|
|
# Release and acceptance deployments set this to an immutable tag or digest. The local
|
|
# fallback remains source-buildable but never floats to `latest`.
|
|
image: ${MODELFORGE_RUNTIME_WORKER_IMAGE:-modelforge-runtime-worker:${MODELFORGE_VERSION:-local}}
|
|
build:
|
|
context: .
|
|
dockerfile: runtime-worker/Dockerfile
|
|
args:
|
|
MODELFORGE_VERSION: ${MODELFORGE_VERSION:-0.0.0}
|
|
MODELFORGE_COMMIT: ${MODELFORGE_COMMIT:-}
|
|
MODELFORGE_BUILT_AT: ${MODELFORGE_BUILT_AT:-}
|
|
hostname: ${MODELFORGE_AGENT_HOSTNAME:-modelforge-node-agent}-runtime
|
|
env_file:
|
|
- path: .env
|
|
required: false
|
|
environment:
|
|
MODELFORGE_RUNTIME_WORKER_CONTROL_PLANE_URL: ${MODELFORGE_AGENT_CONTROL_PLANE_URL:-https://modelforge.example.invalid}
|
|
MODELFORGE_RUNTIME_WORKER_CREDENTIAL_FILE: /run/node-state/node-credential
|
|
MODELFORGE_RUNTIME_WORKER_ARTIFACT_ROOT: /models/model-registry
|
|
MODELFORGE_RUNTIME_WORKER_TLS_VERIFY: ${MODELFORGE_AGENT_TLS_VERIFY:-true}
|
|
MODELFORGE_RUNTIME_WORKER_POLL_INTERVAL_SECONDS: ${MODELFORGE_RUNTIME_WORKER_POLL_INTERVAL_SECONDS:-1}
|
|
HF_HUB_OFFLINE: "1"
|
|
TRANSFORMERS_OFFLINE: "1"
|
|
HF_DATASETS_OFFLINE: "1"
|
|
TOKENIZERS_PARALLELISM: "false"
|
|
volumes:
|
|
- ${MODELFORGE_AGENT_STATE_VOLUME:-node-agent-state}:/run/node-state:ro
|
|
- ${MODELFORGE_AGENT_ARTIFACT_VOLUME:-model-artifacts}:/models:ro
|
|
read_only: true
|
|
tmpfs:
|
|
- /tmp:size=512m,mode=1777
|
|
cap_drop:
|
|
- ALL
|
|
security_opt:
|
|
- no-new-privileges:true
|
|
pids_limit: 512
|
|
restart: unless-stopped
|
|
deploy:
|
|
resources:
|
|
reservations:
|
|
devices:
|
|
- driver: nvidia
|
|
count: 1
|
|
capabilities: [gpu]
|
|
|
|
volumes:
|
|
node-agent-state:
|
|
model-artifacts:
|