Laya Multilingual

Use it to classify and extract structured answers from multilingual messages, tickets, emails, and other text.

Availability

Hugging Face checkpoint; install with `pip install laya`

Pricing model

$0 self-hosted

Deployment

Can run on CPU or compatible GPU; supports local checkpoint directories

Privacy

Subsequent predictions stay local

Licence

Apache 2.0

• The multilingual checkpoint is available in the repository's multilingual subfolder. ^[1]

  Name the checkpoint with `model="multilingual"`, since long mostly-English text would otherwise route to the English checkpoint.

• Version 0.4.0 is available as the laya-coreml Python package on PyPI.
• The runtime supports batch and long-text prediction, hooks, structured decisions, calibration, HTTP/MCP, evaluation, and Router behavior. ^[2]

  prediction, hooks, structured decisions, calibration, HTTP/MCP, evaluation and Router behavior now

• After downloading a model, users can run inference offline on Apple Silicon.
• The runtime returns probabilities for choice, ordinal-score, and boolean questions without autoregressive decoding. ^[2]

  Laya returns probabilities for **choice**, ordinal **score**, and boolean **noul**

• Subsequent predictions stay local. ^[2]

  models download before initialization; subsequent predictions stay local.

• The project is an independent port and is not an official Convai Innovations or Apple release. ^[2]

  Not an official Convai Innovations or Apple release.

• DeepOpen evaluates predefined questions over text or JSON and returns labels, scores, and probabilities.
• The SDK provides choice, score, and noul decision types. ^[2]
• Installation instructions install the package directly from the GitHub repository.
• The quick start supports CPU use and compatible GPU environments.
• Predictions can be wrong, and confidence scores are not a guarantee of correctness.
• Self-hosting avoids per-token API fees but still incurs hardware, electricity, and operating costs. ^[3]

  Self-hosting avoids per-token API fees but still incurs hardware, electricity, and operating costs.

• The model covers more than 100 languages. ^[4]
• It returns typed answers with probabilities in a single forward pass. ^[4]

  (text, email, ticket, or JSON) and **typed questions**; it returns typed answers with

• The checkpoint supports up to 8,192 tokens when `max_len=8192` is set. ^[4]

  > **Long documents: `laya-multilingual` reads up to 8,192 tokens.** It ships with a 1,024-token limit that cuts long documents off, so pass `max_len=8192` for them:

• The checkpoint is available at `convaiinnovations/laya-multilingual` on Hugging Face. ^[4]

  Part of the [Laya family](https://huggingface.co/convaiinnovations/laya) — **use this checkpoint

• The model is systematically over-confident and ships uncalibrated. ^[4]
• The listed licence is Apache 2.0. ^[4]