| --- |
| title: Laya Demo |
| emoji: 🎯 |
| colorFrom: indigo |
| colorTo: blue |
| sdk: gradio |
| app_file: app.py |
| pinned: false |
| license: apache-2.0 |
| short_description: Fast System 1 decisions with calibrated probabilities |
| --- |
| |
| # Laya Demo |
|
|
| Laya is a fast System 1 decision engine: send a **state** and **typed questions**, get typed answers with probabilities and a |
| confidence score. It never generates text, so there is nothing to parse and nothing to hallucinate. |
|
|
| | type | question | answer | |
| |---|---|---| |
| | `choice` | which of these options? | the option, a probability per option, confidence | |
| | `score` | where on this rubric? | a position along your levels, probabilities, confidence | |
| | `noul` | is this true? | the probability that it is | |
|
|
| The tabs are the patterns people use most: support triage, email and phishing, LLM guardrails, RAG passage filtering, |
| moderation, model routing, and a free-form playground. Every tab asks all of its questions in one pass and then decides |
| what to do with plain code: the thresholds live in the app, not in the model. |
|
|
| ## Multilingual routing |
|
|
| The **Multilingual routing** tab runs `laya.Router`: the script of the incoming text decides |
| which checkpoint answers, *before* the forward pass. |
|
|
| | checkpoint | best at | |
| |---|---| |
| | [`laya`](https://huggingface.co/convaiinnovations/laya) | English | |
| | [`laya-multilingual`](https://huggingface.co/convaiinnovations/laya-multilingual) | 100+ languages, ~2x faster | |
| | [`laya-typed-decisions`](https://huggingface.co/convaiinnovations/laya-typed-decisions) | the typed-decisions workflows | |
|
|
| Routing is decided up front because the model's own confidence gives no warning when it cannot |
| read the input. Across 51 languages on MASSIVE intent (20 options, random = 0.050) the English |
| checkpoint macro-averages **0.227** and clears 3x random on **23 of 51** languages, while |
| reporting 0.89-0.99 confidence throughout — on Khmer it scores **0.000 accuracy at 0.952 |
| confidence**. The multilingual checkpoint reaches **0.366** and clears 3x random on **45 of 51**. |
|
|
| ```python |
| from laya import Router |
| router = Router() |
| router.predict({"body": "I was charged twice"}, questions) # -> laya |
| router.predict({"body": "मुझसे दो बार शुल्क लिया गया"}, questions) # -> laya-multilingual |
| ``` |
|
|
| **This is a preview checkpoint** (421M parameters, trained with pure reinforcement learning against proper scoring rules on |
| public datasets). It is strong on routing, classification, moderation and guardrails, weaker on rubric scores, and it saw |
| no email data, so treat the email tab as generalisation rather than a trained skill. |
|
|
| Built by [Convai Innovations](https://huggingface.co/convaiinnovations). |
|
|
