VastAI

skyward.VastAI

Vast.ai GPU marketplace provider.

Parameters:

Name	Type	Description	Default
api_key	str | None	Vast.ai API key. None reads from VASTAI_API_KEY env var.	None
min_reliability	float	Minimum machine reliability score (0.0-1.0).	0.95
verified_only	bool	Only use verified machines.	True
min_cuda	float	Minimum CUDA version.	12.0
geolocation	str | None	Geographic filter (e.g., "US", "EU").	None
bid_multiplier	float	Multiplier for spot bid price above minimum.	1.2
docker_image	DockerImage | None	Custom Docker image. None uses Skyward's default.	None
disk_gb	int	Disk space in GB.	100
use_overlay	bool	Enable overlay networking for multi-node clusters.	True
overlay_timeout	int	Seconds to wait for overlay network setup.	120
require_direct_port	bool	Require direct SSH port (no NAT).	False
min_inet_down	float | None	Minimum download speed in Mbps. None disables filter.	None
min_inet_up	float | None	Minimum upload speed in Mbps. None disables filter.	None

Examples:

>>> sky.VastAI(min_reliability=0.98, verified_only=True)

type property

__init__(*, api_key=None, min_reliability=0.95, verified_only=True, min_cuda=12.0, geolocation=None, bid_multiplier=1.2, instance_timeout=300, request_timeout=30, docker_image=None, disk_gb=100, use_overlay=True, overlay_timeout=120, require_direct_port=False, min_inet_down=None, min_inet_up=None)

create_provider() async

default_options()