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
fix(rust): correct the batch changelog entry and require Python 3.12 for the confidence probe
kevin-gatimu1b1e7d3cd4d9
diff --git a/laya-rust/CHANGELOG.md b/laya-rust/CHANGELOG.mdindex 06982bf3..d4ffbb35 100644--- a/laya-rust/CHANGELOG.md+++ b/laya-rust/CHANGELOG.md@@ -53,10 +53,10 @@ regenerated from laya (Python) 0.4.14 for the supported paths (see the known gaprequests by checkpoint and then by question schema, restores request order, attaches eachrequest's `routing`, and honours each request's `model`, `task`, `lang` and `lang_guess`;`route_batch` loads nothing. New `BatchOptions`, `BatchRequest` (both `#[non_exhaustive]`, with- chained setters, so hooks, `min_confidence`, `lang` and per-request token budgets can be added+ chained setters, so hooks, `lang` and per-request token budgets can be addedlater), `LayaBatchPredictor` with the provided method `LayaPredictor::as_batch_predictor` (adefault of `None`, so existing implementors keep compiling) and `LayaError::InvalidState`, raised- for a `null` state as Python raises `TypeError`. Hooks, `min_confidence`, per-call `lang` and+ for a `null` state as Python raises `TypeError`. Hooks, per-call `lang` and`max_len` / `head_max_len` and the router's `diagnostics` report are not part of it. `LayaError`is not `#[non_exhaustive]`, so an exhaustive `match` on it outside the crate needs an arm for thenew variant.diff --git a/laya-rust/tools/dump_routing_golden.py b/laya-rust/tools/dump_routing_golden.pyindex ea0505f3..856e73f5 100644--- a/laya-rust/tools/dump_routing_golden.py+++ b/laya-rust/tools/dump_routing_golden.py@@ -250,6 +250,10 @@ def _gate(answers, threshold):def build_confidence_probe():+ # jev_confidence sums floats; CPython 3.12 made sum() compensated, and the SDKs reproduce that.+ if sys.version_info < (3, 12):+ sys.exit("confidence_probe.json must be recorded with Python 3.12 or later "+ "(sum() of floats is compensated from 3.12 on), got %d.%d" % sys.version_info[:2])scalars = [dict(value=v, **_checked(confidence.check_min_confidence, v)) for v in cc.SCALAR_THRESHOLDS]maps = []