Skip to content

armgit / laya

Public Imported from GitHub
0

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.

docs(hooks): align calibration input and fixture requirements

xpf16469bf1548
diff --git a/docs/hooks/api.md b/docs/hooks/api.md
index 11d5ae9c..e02758f3 100644
--- a/docs/hooks/api.md
+++ b/docs/hooks/api.md
@@ -178,11 +178,14 @@ Collect `soft_stability` with `accept=None, escalate=None` on labelled evaluatio
then pass those numeric values and correctness labels to `laya.hooks.calibrate_reliability`.
Only fit or replay probed, numeric scores; exclude unprobed reports with `soft_stability=None`
and keep their `decision=None`.
-The fitter requires equal, nonempty score and label sequences. Scores and both accuracy
-targets must be finite Python `int`/`float` values in `[0, 1]` (excluding `bool`); invalid
-lengths or values raise `ValueError`. Correctness labels must be Python `bool` values,
-otherwise it raises `TypeError`. It does not drop unmatched rows or coerce labels such
-as `0`/`1` or `"false"`.
+The fitter requires equal, nonempty, sized score and label sequences; generators raise
+`TypeError`. Scores and both accuracy targets must be finite `numbers.Real` values in
+`[0, 1]`, including NumPy scalars such as `numpy.float32` and `numpy.int64`, but excluding
+`bool` and `numpy.bool_`; invalid lengths or values raise `ValueError`. Correctness labels
+must be `bool` or `numpy.bool_` values (as produced by comparing NumPy arrays), otherwise
+it raises `TypeError`. It does not drop unmatched rows or coerce labels such as `0`/`1`
+or `"false"`. Returned cutoffs are plain Python `float` values, including when fitting
+NumPy inputs, so they can be saved in JSON.
It chooses cutoffs from the distinct input scores at their supplied precision; equal scores
stay in the same group. `ACCEPT` includes scores equal to `accept`, while `ESCALATE` requires
a score strictly below `escalate`. If no cutoff meets the requested accuracy, `accept=1.01`
diff --git a/laya-ts/scripts/option_stability_fixtures.py b/laya-ts/scripts/option_stability_fixtures.py
index 004c2169..b0d3d1ab 100644
--- a/laya-ts/scripts/option_stability_fixtures.py
+++ b/laya-ts/scripts/option_stability_fixtures.py
@@ -4,11 +4,11 @@
python laya-ts/scripts/option_stability_fixtures.py --python-root REFERENCE --check
REFERENCE must contain laya/hooks.py and laya/common.py. The committed oracle records
-its actual source hashes and Git provenance; it uses the rename semantics from #1286
-and the raw score/report/calibration contract from #1289. It is a local combination
-of those changes until they land, not a claim that they are already on main.
-Only the standard library is needed. The production render functions are loaded
-unchanged from their AST so generating fixtures does not import torch or a model.
+its original source hashes and Git provenance from a local combination of the #1286
+rename semantics and the raw score/report/calibration contract from #1289.
+NumPy is required by the production hooks. The production render functions are
+loaded unchanged from their AST so generating fixtures does not import torch or
+a model.
"""
from __future__ import annotations