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(hooks): link every state of a call in the nested-call pattern
Aashish94e4c7dd7048
diff --git a/docs/hooks/tracing.md b/docs/hooks/tracing.mdindex d01c4434..c63b8e21 100644--- a/docs/hooks/tracing.md+++ b/docs/hooks/tracing.md@@ -160,10 +160,14 @@ independent unless you link them yourself. Capture the parent id and pass it alo```pythondef enrich(ctx):- child = enricher.predict(ctx.states[0], EXTRA_QUESTIONS)- record_child_span(parent_run_id=ctx.run_id, child_run_id=child.get("run_id"))+ for state in ctx.states: # a hook sees every state of the call+ child = enricher.predict(state, EXTRA_QUESTIONS)+ record_child_span(parent_run_id=ctx.run_id, child_run_id=child.get("run_id"))```+One parent `run_id`, one child call per state. A body that reads `ctx.states[0]` links the first+decision of a batch and silently drops the rest.+Guard against recursion (see [anti-patterns](patterns.md#recursive-predict)); the easiest guardis a separate `enricher` agent with no hooks.diff --git a/tests/test_hooks_api.py b/tests/test_hooks_api.pyindex 635b3e21..59582f32 100644--- a/tests/test_hooks_api.py+++ b/tests/test_hooks_api.py@@ -353,6 +353,42 @@ check("patterns.md/enrich stays aligned with states",enrich(ctx)check("patterns.md/enrich still guards the recursion", len(ctx.results), len(BATCH))++class ChildCaller(Enricher):+ """The nested-call page's stand-in enricher: it remembers the child calls it is asked to make."""++ def __init__(self):+ self.calls = []++ def predict(self, state, questions):+ result = Enricher.predict(self, state, questions)+ result["run_id"] = "child-%d" % (len(self.calls) + 1)+ self.calls.append(state)+ return result+++SPAN_LINKS, CHILD = [], ChildCaller()+nested = taught("docs/hooks/tracing.md", "enrich", after="## Nested calls",+ enricher=CHILD, EXTRA_QUESTIONS=ASK,+ record_child_span=lambda **kw: SPAN_LINKS.append(kw))+check_true("docs/hooks/tracing.md/enrich reads every state, not index 0",+ "[0]" not in nested.__taught_body__,+ "the body still indexes [0]: %r" % nested.__taught_body__)+nested_ctx = taught_ctx()+nested(nested_ctx)+check("docs/hooks/tracing.md/enrich makes one child call per state", CHILD.calls, BATCH)+check("docs/hooks/tracing.md/enrich links every child to the parent run_id",+ [link.get("parent_run_id") for link in SPAN_LINKS], [nested_ctx.run_id] * len(BATCH))+check("docs/hooks/tracing.md/enrich records each child's own run_id",+ [link.get("child_run_id") for link in SPAN_LINKS],+ ["child-%d" % i for i in range(1, len(BATCH) + 1)])+SPAN_LINKS.clear()+CHILD.calls.clear()+one_child = taught_ctx(states=[BATCH[1]], confidence=[0.4])+nested(one_child)+check("docs/hooks/tracing.md/enrich on one state still links", SPAN_LINKS,+ [{"parent_run_id": one_child.run_id, "child_run_id": "child-1"}])+# ------------------------------------------------ taught token-budget bodies size the whole call## A start hook's ctx.head_max_len / ctx.max_len REPLACE the budgets in force (laya/agent.py), and