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(serve): align head_max_len handling with core library behavior
Swatantra Yadav0cf633460969
diff --git a/laya/serve.py b/laya/serve.pyindex 989f0e26..2c6f19d0 100644--- a/laya/serve.py+++ b/laya/serve.py@@ -394,11 +394,6 @@ def create_app(router: Optional[Any] = None):max_budget_cap = _resolve_max_token_budget()max_len = _validate_budget_param(body, "max_len", max_budget_cap)head_max_len = _validate_budget_param(body, "head_max_len", max_budget_cap)- if max_len is not None and head_max_len is not None and head_max_len >= max_len:- raise HTTPException(- status_code=422,- detail="head_max_len (%d) must be less than max_len (%d)" % (head_max_len, max_len),- )predict_kwargs = {}if max_len is not None:predict_kwargs["max_len"] = max_lendiff --git a/tests/test_serve.py b/tests/test_serve.pyindex 426632e9..e152f7a2 100644--- a/tests/test_serve.py+++ b/tests/test_serve.py@@ -564,11 +564,13 @@ def test_token_budget_exceeds_server_cap(monkeypatch):assert "exceeds server limit" in r.json()["detail"]-def test_token_budget_head_max_len_exceeds_max_len(monkeypatch):+def test_token_budget_head_max_len_equal_to_max_len(monkeypatch):+ """Core accepts head_max_len == max_len; serve forwards both without artificial restriction."""client, fake = _client(monkeypatch)r = client.post("/v1/systemone", json={**REQ, "max_len": 512, "head_max_len": 512})- assert r.status_code == 422- assert "must be less than max_len" in r.json()["detail"]+ assert r.status_code == 200+ assert fake.calls[0]["max_len"] == 512+ assert fake.calls[0]["head_max_len"] == 512def test_token_budget_env_cap_override(monkeypatch):