Laya

What it's good at

• Runs locally, so your data never leaves your server. Air-gapped deployment works.
• You can fine-tune it on your own labels.
• No per-call fee.
• Multilingual checkpoint with automatic routing by script.

Limits and caveats

• The base model is close to chance zero-shot: 0.362 on the typed-decisions benchmark against a 0.318 random baseline.
• The headline 0.766 accuracy comes from a checkpoint fine-tuned on that same benchmark's training split.
• Probabilities are over-confident until you fit temperatures on your own data.
• Convai's own README calls it a fast base to specialise, not a zero-shot decision engine.
• Its “6–8× faster than Jev” compares local GPU inference against Jev's network round trip.

Directory entries using Laya