Non-autoregressive System 1 decision engine. Typed choice, score and yes/no decisions over any text in a single forward pass, in 100+ languages, with a router that picks the right checkpoint per request.
Python calibrationclassificationdecision-modelhuggingfacejevmodernbertmultilingualnlppythonpytorchroutingtyped-decisionszero-shot
docs(research): point at the eval harnesses
Emiliano Gandini Outeda04998bc53bfb
diff --git a/research/README.md b/research/README.mdindex 15783662..3bf63af7 100644--- a/research/README.md+++ b/research/README.md@@ -3,6 +3,13 @@Benchmark harnesses and raw results for the Laya checkpoints. This branch is the evidence behindthe numbers quoted in the main README — nothing here is imported by the `laya` package.+## Evaluation++- [`eval/`](eval/README.md): the independent per-language MASSIVE harness (`laya_eval.py`), the per-case+ report behind the published numbers, plus the metamorphic and presentation checks.+- [`evals/`](evals/README.md): the labelled datasets, thresholds and regression gate consumed by+ `laya.evals` / `laya-eval` and the scheduled `evals` workflow.+## Community diagnostics- [Chinese workplace decisions (Feishu-style)](benchmarks/feishu_zh/README.md) — 64 synthetic scenarios, paired recorded Laya/Jev responses, English/Chinese cards, and a model-free audit. [中文入口](benchmarks/feishu_zh/README.zh-CN.md). Start with `python research/benchmarks/feishu_zh/audit.py`; no downloads or API keys required. This is a contributed historical snapshot, separate from the upstream sweeps below.