A Myanmar-language Laya checkpoint trained on Davlan/sib200, with a Gradio demo space from the same author.
GitHubMultilingual
GitHub
71 builds · page 1 of 2
An Arabic Laya checkpoint trained on AmazonScience/massive and facebook/xnli, with a companion laya-ara-rag ranking model from the same author.
GitHubMultilingual
laya-ara: Laya fine-tuned for Arabic NLU on MASSIVE and XNLI
A Laya checkpoint trained on PolyAI's Banking77, the 77-intent banking support dataset, tagged for customer-support intent classification.
GitHubSupport & triage
Laya fine-tuned on Banking77 for customer-support intent classification
AXERA-TECH's conversion of all three Laya checkpoints for its edge AI chips, reported at latency under 70 ms per decision on AX650/AX8850.
GitHubTools & apps
Laya on Axera AX650 and AX8850 edge NPUs, under 70 ms
Google's LiteRT community conversion of the multilingual checkpoint (tflite), tagged for Android text classification. An English LiteRT build is published alongside it.
GitHubMultilingual
Laya multilingual in LiteRT for Android and on-device
Fan-out SUMs, = NULL, UPDATE without WHERE, injection remnants and more, each with intent and schema context. Four typed questions per statement, identical payloads to both models.
GitHubSecurity & fraud
SQL review benchmark: 14 PostgreSQL statements, seven with deliberate bugs
Synthetic Chinese workplace-chat scenarios with frozen inputs, prompts and labels. Reports classification quality, mis-generated tasks and latency: Jev 253 ms vs Laya 151 ms.
GitHubMultilingual
Feishu message classification bench: Jev 64/64, Laya 20/64 on 64 frozen scenarios
- Single choice
- Jev 64/64, Laya 20/64
20 real-world Chinese requests × 3 checkpoints, measuring accuracy, latency and whether confidence knows when it is wrong. Includes a Windows/Python 3.14 segfault workaround.
Automotive, embedded, enterprise software and product decisions, three trials each, with questions, expected answers, raw responses and a static HTML report.
GitHubMultilingual
Japanese business-judgment benchmark: Jev 97.6 vs Laya 36.9 on the same 40 questions
- Jev 1.13.0
- 97.6 / 100
- Laya multilingual FP16
- 36.9 / 100
Reproducible zero-shot benchmark on informal Moroccan Darija reviews in Arabic script and Arabizi, positive/neutral/negative. Deterministic seed-42 split, no training.
Postgres FTS, pgvector and RRF first stage; Jev, Voyage, GPT-4.1-mini, GPT-5.6-luna and open-weight Laya as rerankers on two Finnish legal corpora. Recall, MRR, nDCG, latency.
Seven zero-shot extraction and classification families demoed, benchmarked and cross-compared, with sizes, licences and costs in one table.
GitHubDocuments & data
zero-shot-ie-bench: GLiNER 2.5, GLiFormer, GLiClass, Laya, Von and Jev in one web UI
Frozen protocol, raw predictions, calibration and latency for Jev 1.13, PrismNLI-0.4B and Laya. Laya ties Jev on dair-ai/emotion and trails by 10–33 points elsewhere.
GitHubDocuments & data
Zero-shot benchmark: Laya's advertised win over Jev does not reproduce on four datasets
- tweet_topic
- Jev 0.793 vs Laya 0.632
- emotion
- 0.587 vs 0.587 tie
Six classifiers over SST-2, AG News and more, reporting accuracy, macro-F1, ECE, p50/p95 latency, throughput and cost side by side. Laya runs locally on MPS or CPU.
GitHubDocuments & data
jevbench: Jev vs GPT-5-mini, Claude Sonnet 5, fine-tuned DistilBERT, BART NLI and Laya
Matched-budget fine-tunes on Banking77 and five more public workflows (prompt injection, SMS spam, emotion, request routing), plus two Aegis chat-moderation fine-tunes.
GitHubModeration & safety
Jev vs fine-tuned Laya: Banking77 goes from 51.3% to 79.4% with 1,001 examples
- Banking77 Laya
- 51.3% → 79.4%
- Banking77 Jev
- 80.0%
Independent Jev vs Laya measurements plus a cascade: Laya decides first, high confidence executes, low confidence escalates to Jev. At a 0.60 threshold it matched Jev's 78%.
GitHubLLM routing & guardrails
laya-jev-lab: a local-first cascade that matches Jev's accuracy at 1.8x the speed
- Cascade accuracy
- 78%, same as Jev
- Speed
- 1.8x Jev, as reported
MCP server plus HTTP sidecar: token-budget preflight that reports what would be cut, a persisted calibration store, structured errors, and a noul fix (P(true) 0.5 → 1.0).
Anthony Alcaraz's talk demo routes typed decisions through Jev and shows Laya's answers beside them. On the TSMC run Laya rejected every true claim; that's the point.
GitHubDocuments & data
Agentic GraphRAG over S&P 10-Ks with Laya as a shadow decision engine beside Jev
Local 15x15 Gomoku on Apple Silicon. Rules in Python, board in the browser, every AI move one real Laya choice over up to six candidates, probabilities shown. No Gomoku training.
A port of the jev-ultrafast browser agent that decides with laya-mlx instead of the hosted Jev API. DOM snapshot, executor, safety checks and inspector are the originals.
GitHubLLM routing & guardrails
Laya Ultrafast: browser-use's jev-ultrafast agent ported to local Laya on MLX
From the Prompt Engineer 48 video. Both models get the same input and typed questions; only the model differs. Laya needs no key.
One typed question per frame; the other paddle is three lines of arithmetic as a control. Physics in WebAssembly, decisions in native Laya over POST /decide.
The game enumerates placements with their holes, heights and clears; Laya picks one from text. The teacher is a board-evaluation formula, not Jev. Walkthrough video in Japanese.
Deterministic arcade physics, a browser flight recorder and a supervised Laya fine-tune that landed 30 of 30 held-out flights on Apple MLX. 58% less peak memory with 6-bit weights.
GitHubTools & apps
Lunar Laya: Atari-style lander with a CUDA-trained Laya checkpoint that lands 30/30 on MLX
- Held-out landings
- 30 / 30
- Peak memory
- -58% with 6-bit
Native Windows demo with live probabilities, executed moves, inference timing and planner-intervention counts. A deterministic search proposes routes, Laya classifies them.
Unmodified Laya weights, real NES emulation via Stable Retro, text observations from RAM, model-driven jumping. 27.5 s of game time at 16.79 ms median inference.
GitHubTools & apps
Laya clears Super Mario Bros. 3 World 1-1 on a DGX Spark: 207 decisions, frames verified
- Decisions
- 207
- Game time
- 27.50 s
Abhimanyu walks the Mahabharata ring formation to the centre. Each step goes to local Laya as typed questions; unsure or unplayable answers are marked red and corrected.
GitHubTools & apps
Chakravyuha: a browser ring-maze where every move is a Laya decision
Two routers behind one endpoint, measured on 180 labelled requests. Laya answers small/medium/powerful routing in one forward pass, against a GPT-5 nano router.
GitHubLLM routing & guardrails
laya_router: a model router with Laya matches GPT-5 nano's accuracy, 35x faster, free
- vs GPT-5 nano
- same accuracy, 35x faster, as reported
- Requests
- 180
Measures the decision head's attention drop-off on long inputs across 400 documents, then chunks, screens and aggregates, or locates the one passage that matters.
One simulated Swedish villa, four controllers on the same minutes. Laya with a safety clip cut the bill from 985 to 530 SEK at p95 179 ms; a plain tariff EMS reached 385 SEK.
GitHubTools & apps
A public test of Laya on a home energy system: fast enough, but a tariff EMS still wins
- Bill
- 530 SEK vs 985 naive
- Tariff EMS
- 385 SEK
Collects vacancies, filters them against a candidate profile, ranks with the multilingual Laya checkpoint and prints IT and warehouse tables plus a Markdown report.
GitHubDocuments & data
Job search in Denmark: aggregate jobnet.dk and jobindex.dk, rank matches with Laya
Self-hosted console that runs VirusTotal lookups through System One models and evaluates them against each other on a labelled dataset. Prompts and criteria are versioned JSON.
GitHubSecurity & fraud
IOCArena: classify IPs, domains and hashes from VirusTotal data with Jev, Von or Laya
Replayable grid sim with moving workers and forklifts. Laya answers action, collision_risk, path_blocked and needs_operator in one call; collision prevention stays deterministic.
GitHubModeration & safety
Laya Warehouse Safety: a robot picks advance, shift or wait above a deterministic shield
Scores post style locally through a Laya-MLX helper as you scroll, using low-information and formulaic-rhetoric criteria and a conservative consensus score. Abstains when unsure.
GitHubModeration & safety
Slop Finder: a Chromium extension that flags AI-slop style on X, LinkedIn and Reddit
Paste a nervous message and watch tone (choice), formality 1–5 (score) and fight risk (noul) update as you type, 40–110 ms per call. Misses subtle passive-aggression.
GitHubModeration & safety
Before You Send: live tone, formality and fight-risk readings as you type
Companion service that reads Lidarr's manual-import queue, scores every candidate release and track in one pass, and picks the right Single/EP/Album with calibrated probabilities.
GitHubDocuments & data
lidarr-decision-import: resolve Lidarr's stuck imports with Laya instead of an LLM
Describe a past conversation in plain language and Chat Seek finds it across local agent histories, reranks with the receptron Laya runtime, and offers a Resume action.
Allen Porter's native Assist conversation agent scores intents and entities in one Laya pass, runs device commands locally and escalates open-ended queries to a fallback LLM.
GitHubTools & apps
Laya for Home Assistant: a fully local Assist conversation agent
Swift package that reads a form in a running Mac app through the Accessibility API, asks an on-device model what belongs in each field, and types it in. ~1 ms per decision.
GitHubTools & apps
FluidUse: local computer use on Apple Silicon with Laya and CUA-S1-FORMS
Local decider for the browser-agent loop: hand it a numbered table of page controls and it picks the operation and element. Playwright/CDP driver, speaks /v1/systemone.
GitHubLLM routing & guardrails