Files
ray-project--ray/release/serve_tests/compute_tpl_gpu_node_gce.yaml
2026-07-13 13:17:40 +08:00

18 lines
461 B
YAML

cloud: {{env["ANYSCALE_CLOUD_NAME"]}}
zones:
- us-west1-b
head_node:
instance_type: n1-standard-16-nvidia-tesla-t4-1 # g4dn.4xlarge
# GPU head — explicitly expose the instance's natural 16 vCPU + 1 GPU so the
# test driver (on the head) can actually run GPU workloads.
resources:
CPU: 16
GPU: 1
worker_nodes:
- instance_type: n2-standard-16 # m5.4xlarge
min_nodes: 0
max_nodes: 1
market_type: ON_DEMAND