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
release: 0.3.11
Nandakishor1e28ac20c089
diff --git a/README.md b/README.mdindex 593bb96b..f0c83e51 100644--- a/README.md+++ b/README.md@@ -34,6 +34,16 @@ Three checkpoints, and a `Router` that picks between them per request:| [`laya-multilingual`](https://huggingface.co/convaiinnovations/laya-multilingual) | mmBERT-base | 322M | 1024 | 100+ languages, 2x faster || [`laya-typed-decisions`](https://huggingface.co/convaiinnovations/laya-typed-decisions) | ModernBERT-large | 421M | 1024 | the typed-decisions workflows |+### What's new in 0.3.11++* **Routed batches.** `Router.predict_batch(requests)` routes each request, groups them by checkpoint and question set, and scores each group in shared forward passes, with answers identical to one `predict` call per request. Each request can set its own `model`, `task`, `lang` or `lang_guess`. See [Heterogeneous routed batches](#heterogeneous-routed-batches).+* **Prediction hooks.** Opt-in hooks run around every decision on `Agent`, `Router` and `ONNXAgent`, to audit, trace, redact, cache or gate results. With no hooks set, answers are identical to before. See [Prediction Hooks](#prediction-hooks).+* **transformers 4.x and Apple GPUs.** Checkpoints re-saved by transformers 5 now load with the right RoPE settings on transformers 4.x, and `predict()` no longer crashes on MPS builds without an autocast backend.+* **Stricter `noul` questions.** A `noul` `criteria` dict keyed anything other than `true`/`false` is now rejected with a clear message instead of being silently replaced by the defaults. Use `labels` to change the wording.+* **Safer HTTP server.** Timing-safe API key checks, request size and question limits (413), a 400 for malformed JSON, errors that do not leak paths, and a validated port. Docker Compose binds `laya-serve` to loopback by default and adds a healthcheck.+* **More hardware and docs.** Native ARM64 and DGX Spark container builds, a documentation site built from `docs/`, and a local web GUI demo under `examples/`.+* **Smaller fixes.** Plain-ASCII German routes to the multilingual checkpoint, very large e-mails and states are bounded before regex work, `ONNXAgent` validates questions like `Agent`, an empty `HF_TOKEN` no longer breaks downloads, the tokenizer config is written atomically, and the `langchain` extra installs `langgraph`.+### What's new in 0.3.100.3.10 changes only this README; its code is the same as 0.3.9. Everything below is new since 0.3.6. `pip install -U laya` for all of it; the checkpoints are unchanged.diff --git a/laya/__init__.py b/laya/__init__.pyindex 330016e8..cb41e415 100644--- a/laya/__init__.py+++ b/laya/__init__.py@@ -13,7 +13,7 @@ from .presets import ()from .router import DEFAULT_MODELS, RouteDecision, Router-__version__ = "0.3.10"+__version__ = "0.3.11"# Routing, language detection and email cleaning are pure Python. The torch-backed names are# resolved lazily so that `import laya` -- and therefore `from laya import Router` ordiff --git a/pyproject.toml b/pyproject.tomlindex f4f2808b..67e7094d 100644--- a/pyproject.toml+++ b/pyproject.toml@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"[project]name = "laya"-version = "0.3.10"+version = "0.3.11"description = "Fast, non-autoregressive System 1 decision engine with calibrated probabilities"readme = "README.md"requires-python = ">=3.10"