Code language ID — which language is this snippet?

Why it matters

Detecting a snippet's language — from the code itself, not a filename — is the quiet machinery behind syntax highlighting, gist tagging and code search. Doing it on-device means an editor or docs site can highlight a pasted snippet without shipping the code anywhere.

• Syntax highlighting — pick the right grammar for a pasted, unlabelled snippet.
• Snippet & gist tagging — auto-label code blocks in notes, wikis or chat.
• Code search & routing — filter or route snippets by language before indexing.
• Docs & education — flag when a code block's language tag doesn't match its contents.

It knows the six CodeSearchNet languages (Go, Java, JavaScript, PHP, Python, Ruby) and reads syntax, so a very short or ambiguous snippet — a one-liner, or code that looks alike across languages — can be misread (a bare Go one-liner may score as JavaScript). The confidence bars make that visible, and a few real lines are usually all it needs. It will always pick the closest of its six classes, even for a language it doesn't know.

How the API works

It's one text-classification call on the snippet:

import { pipeline } from "@huggingface/transformers";

const id = await pipeline("text-classification",
  "onnx-community/CodeBERTa-language-id-ONNX");
const out = await id(codeSnippet, { top_k: 6 });
// → [{ label: "python", score: 0.99 }, … ]

CodeBERTa tokenises the code with a byte-level BPE trained on source and scores the six languages. It runs in a Web Worker so typing stays smooth.

References

• Model card — CodeBERTa language ID (ONNX build)
• Original model — huggingface/CodeBERTa-language-id (Apache-2.0)
• CodeSearchNet — the six-language corpus
• Transformers.js documentation