Open-source Laya classifies inputs in a single 33ms forward pass across 100+ languages
ssh4net · x · 2026-09-19
Convai Innovations released Laya, a family of multilingual, non-autoregressive System 1 decision models on Hugging Face.
- Given a state (text, email, ticket, or JSON) and typed questions, it returns typed answers with probabilities in a single forward pass (33ms) across 100+ languages
- Trained with RLCD (reinforcement learning against strictly proper scoring rules), so reporting honest probabilities is the only way to maximize reward; it never generates text, so nothing to parse and nothing to hallucinate
- Three checkpoints: English (ModernBERT-large, 421M), multilingual (mmBERT-base, 322M, 2x faster), and typed-decisions; Apache-2.0 licensed for commercial use
- Designed for classification, routing, scoring, and moderation/guardrail workflows
Related event: Open-Source Decision Model Laya Runs in 33ms, Supports 100+ Languages(3 posts)→
More from Models
- JevBench v1.2 released: Jev 1.13 holds the lead at 75.3, open 4B models close behind — airesearch12 · 2026-09-19
- Fine print: $3,800 per task and still no paradigm-shifting breakthroughs over humans — seanwbren · 2026-09-19
- Jev fills the general-purpose classifier gap: labels in code, no extra LLM call — alexcovo_eth · 2026-09-19
- Google Trends: ChatGPT, Claude and Gemini dominate LLM keyword searches — NickPassig · 2026-09-19
- TipTour goes open-source: local CoreML-driven Mac pointer agent clicks in ~90ms without screenshots — alexcovo_eth · 2026-09-19
- Dev hides AI thinking notes to stop spoilers as GLM-5.3 Flash stays on cost Pareto frontier in chess app — MikePFrank · 2026-09-19