vllm-fly/fly.toml

28 lines
414 B
TOML
Raw Permalink Normal View History

2024-05-21 02:28:10 +02:00
app = "vllm-demo"
primary_region = "ord"
[experimental]
entrypoint = "/start.sh"
[build]
2024-10-03 16:20:34 +02:00
image = "vllm/vllm-openai:latest"
2024-05-21 02:28:10 +02:00
[http_service]
internal_port = 8000
force_https = true
[[vm]]
size = 'l40s'
gpus = 1
[[files]]
guest_path = "/start.sh"
local_path = "start.sh"
2024-10-03 16:25:41 +02:00
[[services]]
internal_port = 8000
protocol = "tcp"
auto_stop_machines = "stop"
auto_start_machines = true
min_machines_running = 0