relay-laya

Trained Laya model used by the laya/auto router of Relay, a coding agent harness. It answers 18 typed questions about a coding request (task type, complexity, scope, risk, recommended capability tier and reasoning effort, agent role, validation level, tools needed, security sensitivity) so Relay can pick the cheapest model likely to succeed.

Relay downloads these files on first use (/laya setup) and checks each one against a sha256 pinned in its package. You normally do not need to download them by hand.

  • Base: multilingual Laya (mmBERT-base), fine-tuned for 4 rounds.
  • Training data: 1100 synthetic, template-generated exercises. Accuracy on its held-out exercises (95-100% per question) overstates real-world accuracy, because they come from the same templates. Expect lower accuracy on real requests until it is retrained with data from real use.
  • Requires the laya Python package (0.3.27) to load.
Downloads last month
16
Safetensors
Model size
0.3B params
Tensor type
F16
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support