Detect Text Language (HuggingFace)

@huggingface1 VersionDetect

Identifies the language of a String (or [String]) with a fine-tuned HuggingFace classifier and emits the model's own language code by output slot: a top-1 String, a per-segment [String], a Classification, or a [Classification].

How it fits

Typical backends

Route transcribed speech to a language tag for downstream branching.

Detect the source language of incoming text before translating it.

Caveats

Parameter interactionmodel and device are hot-swappable at runtime; max_input_tokens, batch_size, and top_k are re-read every message.
I/O contractThe emitted code is the model's own id2label label; the default papluca/xlm-roberta-base-language-detection covers 20 languages as ISO 639-1 codes (e.g. 'en', 'fr', 'zh'). Swapping model changes the label set and code convention.
Parameter interactiontop_k applies only when a single String feeds a [Classification] output, returning that text's ranked languages; on every batch arm and every single-best arm the top-1 language is emitted.
AccuracyTexts longer than max_input_tokens are truncated to the model's max_position_embeddings before detection; language id stays reliable on short prefixes, so a long document does not need chunking.
FallbackEmpty or whitespace-only input (or an all-blank list) returns the empty form of the resolved arm without running the model; a blank entry in a [String] keeps its index with an empty-label verdict so the batch arms stay aligned.

Versions

90340f33latestdefault
Automated release
6/26/2026