laya-multilingual: 100+ Languages, mmBERT-base 322M, Benchmarks & Limits

Laya's multilingual checkpoint on Hugging Face: mmBERT-base, 322M parameters, 1024-token context, 100+ languages, faster than the English model, plus calibration and known limits.

Last updated: Sep 24, 2026

convaiinnovations/laya-multilingual is the Laya checkpoint for anything that is not English. It is also the faster of the two general checkpoints.

Hugging Faceconvaiinnovations/laya-multilingual
EncodermmBERT-base (22 layers, hidden 768, 256k vocabulary)
Parameters322M (307M encoder + decision head)
Default context1,024 tokens (256 for question and options); encoder supports up to 8,192
Download size~647 MB
Languages100+ (model card lists 51 evaluated languages, including Chinese, Japanese, Korean, Arabic, Hindi)
LicenseApache 2.0

Sources: model card and upstream README. Figures are upstream measurements unless noted.

Load it

import laya

agent = laya.load("convaiinnovations/laya-multilingual")
result = agent.predict(
    {"body": "二重に請求されました。返金してください。"},
    {"department": {"type": "choice",
                    "instructions": "Which team should handle this?",
                    "criteria": {"billing": "invoices, payments, refunds",
                                 "technical": "bugs and outages"}}},
)

With the Router, non-English input reaches this checkpoint automatically. The routing decision is made from the script and language before the forward pass. If most of your traffic is not English, set Router(default="multilingual") so very short, ambiguous Latin-script text also lands here, or pass your own language detector with lang_guess=.

Why it exists

Across all 51 MASSIVE languages (intent classification, 20 options, random = 0.050):

laya (English)laya-multilingual
Macro accuracy0.2270.366
Macro ECE (lower is better)0.7330.387
Languages clearing 3x random23 / 5145 / 51
XNLI, 14 non-English languages0.5210.731

Per-language examples from the model card: Arabic 0.110 → 0.400, Hindi 0.100 → 0.387, Korean 0.110 → 0.490, Turkish 0.140 → 0.437.

Speed (Tesla T4)

Questions per calllayalaya-multilingual
139.5 ms32.8 ms
10158.6 ms72.3 ms
50771 ms337 ms

Upstream reports 103–332 questions per second batched on one T4.

Known limits

  • Ships uncalibrated. Temperatures are all 1.0 and the model is systematically over-confident. Refitting on held-out data moves mean ECE from 0.314 to 0.106. Do this before trusting its probabilities; see Fine-tune Laya.
  • Weaker on English than the English checkpoint (0.619 vs 0.684 macro across English suites). Route rather than replace.
  • Low-resource languages are weak: Swahili 0.210, Tamil 0.250, Amharic 0.110.
  • Position bias on score questions: it rarely picks the first-listed level, in any language (#131). For English score questions use the English checkpoint; for others validate first.
  • noul can under-report "true" (#156). A two-option choice with neutral A/B keys is a useful cross-check.
  • Near chance on typed-decisions zero-shot (0.342). Fine-tune for a specific workflow.
  • Keep choice questions under ~20 options.

For Chinese-language workloads

The checkpoint's language list includes Chinese (zh), and CJK text containing Latin brand names is routed here. Upstream also links a community diagnostic of Chinese workplace decisions. As with every language, measure accuracy and calibration on your own data first.

Compare the family

CheckpointEncoderParamsContextUse it for
layaModernBERT-large421M512English
laya-multilingualmmBERT-base322M1024100+ languages
laya-typed-decisionsModernBERT-large421M1024four typed-decisions workflows

Back to the model overview.

Last verified: September 24, 2026.