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.

feat(confidence): export jev_confidence, the formula behind the served x_jev_confidence

Aashishb937ce0ed211
diff --git a/docs/http-api.md b/docs/http-api.md
index a93644ab..39b87bc9 100644
--- a/docs/http-api.md
+++ b/docs/http-api.md
@@ -252,6 +252,10 @@ before the question ids are read.
and `score` answers only (a Jev `noul` answer has no confidence). It uses TypeSafe's own
formulas: `(p_max - 1/n)/(1 - 1/n)` for `choice`, and for `score` one minus the probability-weighted
distance from the most likely level, scaled by the same distance for a uniform distribution.
+ The formula is the library's own, not the server's: `laya.jev_confidence(answer)` takes one
+ answer dict -- the shape this route returns and the shape `Agent.predict` returns -- and gives
+ the unrounded number, so a caller running Laya in-process filters on the same quantity the
+ served field carries without an HTTP round trip.
Never compare them against one threshold. When porting from Jev, a threshold carried over from a
Jev deployment gates differently on Laya's entropy `confidence` (#302); read it against
diff --git a/docs/reference/helpers.md b/docs/reference/helpers.md
index 776e36d1..ac6ac50d 100644
--- a/docs/reference/helpers.md
+++ b/docs/reference/helpers.md
@@ -107,6 +107,8 @@ supplies it.
::: laya.common.confidence_from_probs
+::: laya.confidence.jev_confidence
+
::: laya.common.ece_score
::: laya.calibrate.fit_temperatures
diff --git a/laya/__init__.py b/laya/__init__.py
index 915ccf87..33a1f7a9 100644
--- a/laya/__init__.py
+++ b/laya/__init__.py
@@ -63,6 +63,7 @@ _LAZY_ATTRS = {
"flag_low_confidence": (".confidence", "flag_low_confidence"),
"apply_confidence_gate": (".confidence", "apply_confidence_gate"),
"GATE_STATES": (".confidence", "GATE_STATES"),
+ "jev_confidence": (".confidence", "jev_confidence"),
"render_options": (".common", "render_options"),
"QTYPES": (".common", "QTYPES"),
"QTYPE_NAMES": (".common", "QTYPE_NAMES"),
@@ -135,6 +136,7 @@ __all__ = [
"flag_low_confidence",
"apply_confidence_gate",
"GATE_STATES",
+ "jev_confidence",
"render_options",
"QTYPES",
"QTYPE_NAMES",
diff --git a/tests/test_confidence.py b/tests/test_confidence.py
index 8285e6ac..3c621256 100644
--- a/tests/test_confidence.py
+++ b/tests/test_confidence.py
@@ -2296,6 +2296,32 @@ for _name, _answer in (
):
check("jev_confidence/none-for-%s" % _name, jev_confidence(_answer), None)
+# --- the public name and the served field must be one computation (#302) --------------------
+# `/v1/systemone` publishes `x_jev_confidence`, but a caller running the checkpoint in-process
+# gets an answer dict and no HTTP layer, and the other two confidence quantities
+# (`answer_confidence`, `confidence_from_probs`) are already public. So the name ships from
+# `laya`, and these checks pin that the served number is this function's number.
+import laya # noqa: E402
+
+check_true("jev_public/exported", "jev_confidence" in laya.__all__)
+check_true("jev_public/lazy attr resolves to the core function",
+ getattr(laya, "jev_confidence", None) is jev_confidence)
+
+from laya.serve import _add_jev_confidence # noqa: E402
+
+_answers = {
+ "q": _choice(0.88, 0.12),
+ "r": _choice(2, 1), # 1/3, so the served rounding is visible
+ "s": _score(0.0, 0.95, 0.05),
+ "n": {"type": "noul", "noul": 0.9},
+}
+_served = _add_jev_confidence({"answers": copy.deepcopy(_answers)})["answers"]
+for _qid in ("q", "r", "s"):
+ check("jev_public/served %s is the public function rounded" % _qid,
+ _served[_qid]["x_jev_confidence"], round(jev_confidence(_answers[_qid]), 4))
+check_true("jev_public/served noul carries no field",
+ "x_jev_confidence" not in _served["n"], _served["n"].keys())
+
print("\n%d passed, %d failed" % (len(PASS), len(FAIL)))
for f in FAIL:
print(" FAIL", f)
diff --git a/tests/test_hooks_api.py b/tests/test_hooks_api.py
index 9b4b0c6d..8d324b27 100644
--- a/tests/test_hooks_api.py
+++ b/tests/test_hooks_api.py
@@ -244,6 +244,13 @@ check_true("laya.hooks/run_coroutine_sync exists",
check_true("__all__/PINNED_REVISIONS", "PINNED_REVISIONS" in laya.__all__)
check_true("laya.PINNED_REVISIONS exists", hasattr(laya, "PINNED_REVISIONS"))
+# The three confidence quantities are one family, and `x_jev_confidence` is only half of it: the
+# server publishes the third (#302), so a caller running Laya in-process has to be able to compute
+# it from the library. Pin all three as attributes, not just as `__all__` entries.
+for _confidence_name in ("answer_confidence", "confidence_from_probs", "jev_confidence"):
+ check_true("__all__/%s" % _confidence_name, _confidence_name in laya.__all__)
+ check_true("laya.%s exists" % _confidence_name, hasattr(laya, _confidence_name))
+
# BaseHook is the concrete no-op base class; all six events exist and are callable.
for event in HOOK_EVENTS:
check_true("BaseHook/%s callable" % event, callable(getattr(BaseHook, event, None)))