MixLID

Use it to identify language spans in mixed-language documents or transcripts, filter multilingual data, and evaluate language consistency in generated text.

Availability

Published Hugging Face model ID and Hugging Face Space demo

Pricing model

Not established

Deployment

Model can be downloaded and run with Transformers or the local MixLID module

Privacy

Not established

Licence

MIT

• MixLID supports 111 language labels. ^[1]

  It supports **111 language labels** and returns both **language spans** and **one predominant language per input**.

• It provides token-level predictions and a predominant-language prediction for an input. ^[1]

  | Prediction modes | Token-level predictions and predominant input language |

• Token predictions include language labels, character offsets, and scores. ^[1]

  | Output details | Language labels, character offsets, and prediction scores |

• A Hugging Face Space demo is available. ^[1]

  **Demo:** [Hugging Face Space](https://huggingface.co/spaces/hongst/MixLID)

• The published model can be loaded using Transformers or the local MixLID module. ^[1]

  Load the downloaded model directory, or replace `"./MixLID"` with the published Hugging Face model ID:

• Muse Wonder is aimed at document work, workflows, and coding Daily brief 2026-10-12