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(notebooks): align calibration temperature bounds with runtime
Sarthak Pandey5e840dd027e7
diff --git a/README.md b/README.mdindex 1811bade..4aadb0c4 100644--- a/README.md+++ b/README.md@@ -473,7 +473,7 @@ The hint only decides *English or not*: a code whose primary subtag is `en`, `enwire protocol as TypeSafe's hosted Jev API. Laya's answer payload is alreadyschema-identical to what Jev returns (`choice`/`score`/`noul` answers and a`{input_tokens, output_tokens}` usage block), so an existing Jev client — e.g.-the [`hs-jev`](https://github.com/getmissionctrl/hs-jev) Haskell client or the [`marcreichel/laya-php`](https://github.com/marcreichel/laya-php) PHP / Laravel SDK — just+the [`hs-jev`](https://github.com/getmissionctrl/hs-jev) Haskell client — justneeds its `baseUrl` repointed; nothing else changes.```bashdiff --git a/docs/http-api.md b/docs/http-api.mdindex 38478b60..a1bb894a 100644--- a/docs/http-api.md+++ b/docs/http-api.md@@ -1,7 +1,7 @@# HTTP API`laya-serve` exposes Laya over the TypeSafe Jev `/v1/systemone` wire protocol. A client written-against Jev -- `hs-jev`, `typesafe-sdk`, [`marcreichel/laya-php`](https://github.com/marcreichel/laya-php) (PHP / Laravel), or your own -- can point its base URL at this server and+against Jev -- `hs-jev`, `typesafe-sdk`, or your own -- can point its base URL at this server andkeep working: Laya's `predict()` output is already schema-compatible, and the server adds only theHTTP surface: one decision route, a health probe, an optional bearer check and request limits.diff --git a/laya/serve.py b/laya/serve.pyindex 10a46723..ed08d39a 100644--- a/laya/serve.py+++ b/laya/serve.py@@ -3,7 +3,7 @@Laya's ``predict()`` output is already schema-compatible with the Jev decisionAPI -- ``choice`` / ``score`` / ``noul`` answers and a ``{input_tokens,output_tokens}`` usage block -- so a client written against Jev (for example the-`hs-jev` Haskell client or `marcreichel/laya-php` PHP SDK) can point its ``baseUrl`` at this server and keep+`hs-jev` Haskell client) can point its ``baseUrl`` at this server and keepworking unchanged. All this module adds is the HTTP surface Laya itself does notship: a ``POST /v1/systemone`` route, an optional bearer check, and a healthprobe.diff --git a/notebooks/laya_finetune_typed_decisions_2xT4_kaggle.ipynb b/notebooks/laya_finetune_typed_decisions_2xT4_kaggle.ipynbindex 29ae152d..df14a735 100644--- a/notebooks/laya_finetune_typed_decisions_2xT4_kaggle.ipynb+++ b/notebooks/laya_finetune_typed_decisions_2xT4_kaggle.ipynb@@ -4,7 +4,7 @@"cell_type": "markdown","metadata": {},"source": [- "# Fine-Tuning Laya on `LocalLLaMA/typed-decisions` (Kaggle GPU T4 ×2 DDP)\n",+ "# Fine-Tuning Laya on `LocalLLaMA/typed-decisions` (Kaggle GPU T4 \u00d72 DDP)\n","\n","[](https://pypi.org/project/laya/)\n","[](https://huggingface.co/convaiinnovations/laya)\n",@@ -14,7 +14,7 @@"\n","---\n","\n",- "### ⚠️ Kaggle Notebook Settings:\n",+ "### \u26a0\ufe0f Kaggle Notebook Settings:\n","In the right-hand sidebar panel under **Notebook options**:\n","* **Accelerator:** Select **`GPU T4 x2`** (ensure both GPUs are allocated)\n","* **Internet:** Set to **`On`**\n",@@ -431,7 +431,7 @@" dist.destroy_process_group()\n","\n","if __name__ == \"__main__\":\n",- " main()"+ " main()\n"],"execution_count": null,"outputs": []