Laya: Open-Source Model Returns Calibrated Probabilities in 33ms Across 100+ Languages
victormustar · x · 2026-10-06
Laya (Apache 2.0, 5.2k+ likes) is now on Hugging Face: a multilingual, non-autoregressive System 1 decision model. Given a state (text, email, ticket, or JSON) and typed questions, it returns typed answers with mathematically calibrated probabilities in a single forward pass (33 ms) across 100+ languages.\n\n- Training: reinforcement learning against strictly proper scoring rules (RLCD), so reporting honest probabilities is the only way to maximise reward\n- No text generation: nothing to parse, nothing to hallucinate\n- Deployment: pip install laya, with optional extras for HTTP server, MCP server, LangChain/LangGraph, and ONNX Runtime\n- Long documents: the multilingual variant reads up to 8,192 tokens (default 1,024 truncates; pass maxlen=8192). Accuracy holds up to 4,000 tokens (16-18/20 correct) but varies beyond that (8-17/20)
More from coding & agent
- Codex spins its wheels while Claude Opus shows far deeper understanding, user reports — iruletheworldmo · 2026-10-06
- Urlbox MCP Server Lets You Capture Screenshots, PDFs and Markdown via Natural Language — modelcontextprotocol · 2026-10-06
- OpenAI Dots Thread Steps 7-8: Approval Gates and Cost-Tiered Intelligence — FinanceYF5 · 2026-10-06
- OpenAI Dots Thread Steps 5-6: Wire in Business Tools and Schedules — FinanceYF5 · 2026-10-06
- 10-Step Blueprint Turns OpenAI Dots Into a 24/7 Company Operations Layer — FinanceYF5 · 2026-10-06
- ODS offers one-click local AI: auto hardware detection, model setup, agents and plugins — Teknium · 2026-10-06