Advanced inference backends

Detect capabilities and plan experimental multi-GPU profiles.

arka backend capabilities vllm
arka backend capabilities exo --json
arka backend profile vllm --model my-model --gpus 2 --flash-attention
arka backend profile exo --model my-model --gpus 4 --row-split --co-engine ds4

Profiles are preview-only. Use --run to launch. Arka reports capability assumptions and rejects unsupported combinations before starting a process. Backend flags are experimental and must be validated against the installed runtime version and hardware.