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
"""The two states and four questions the sample app asks, as golden cases.
Kept beside the other cases rather than inline in `golden_cases.py`'s literal list because
these mirror a *different* file: `laya-rust/samples/laya-sample/src/main.rs`. If that sample
changes, these must change with it, and a separate module makes that coupling visible.
The Hindi body is written as escapes deliberately. It was extracted from the sample source
programmatically, not retyped: an earlier hand transcription put U+0915 U+094B (ko) where
the sample has U+091A (ca), which silently compared two different sentences and looked like
a parity bug in the port.
"""
SAMPLE_ENGLISH_STATE = {
"from": "[email protected]",
"subject": "Duplicate charge on invoice #4411",
"body": "Hi, we were billed twice for March. Please refund the duplicate today "
"or we will cancel our plan.",
}
SAMPLE_HINDI_STATE = {
"body": "मार्च का बिल "
"दो बार लिया गया, "
"कृपया रिफंड "
"करें।",
}
# Option order is load-bearing: choice labels are positional, so this dict doubles as the
# order fixture for the sample's QuestionSet.
SAMPLE_QUESTIONS = {
"department": {
"type": "choice",
"instructions": "Which department should handle this request?",
"criteria": {
"billing": "invoices, payments, refunds",
"technical": "bugs, outages, system errors",
"sales": "pricing, new contracts",
"other": "everything else",
},
},
"urgency": {
"type": "score",
"instructions": "How urgent is this request?",
"criteria": ["not urgent", "soon", "critical deadline or blocking issue"],
},
"churn_risk": {
"type": "noul",
"instructions": "Does the user threaten to cancel or leave?",
},
"refund_requested": {
"type": "noul",
"instructions": "Does the user explicitly request a refund?",
},
}
SAMPLE_CASES = [
# The sample's English run. Dict state, so this also pins serialize_state's dialect on
# the exact payload a user sees printed on their first run of the SDK.
{
"name": "sample_app_english",
"state": SAMPLE_ENGLISH_STATE,
"questions": SAMPLE_QUESTIONS,
},
# The sample's Hindi run: the same questions against Devanagari, which is the claim that
# the multilingual checkpoint needs no router. Byte-fallback territory for the tokenizer.
{
"name": "sample_app_hindi",
"state": SAMPLE_HINDI_STATE,
"questions": SAMPLE_QUESTIONS,
},
]