SNIP (Small N-gram Identifier for Pastes)

SNIP is a small browser-local classifier for pasted text and source snippets.

SNIP predicts a likely syntax or text label for pasted text, snippets, logs, configuration files, and text-like source files. It is designed for browser-local inference in applications where sending pasted content to a server is undesirable. It aims to be quick, small in size, and fairly accurate.

• Website/demo: https://snip.wesring.com
• GitHub: https://github.com/wesr/snip
• Hugging Face: https://huggingface.co/wesringml/snip
• npm: https://www.npmjs.com/package/@wesr/snip

• model/snip_model.json - source model weights used to generate the embedded runtime model
• src/snip.ts - dependency-free TypeScript source runtime
• dist/snip.js - compiled JavaScript runtime
• dist/snip.d.ts - TypeScript declarations
• examples/node-example.mjs - minimal Node usage example
• docs/index.html - browser demo page
• metrics/ - evaluation metrics copied from the release candidate
• MODEL_CARD.md - concise model card
• REPORT.md - technical report

Install the package:

npm install @wesr/snip

import { classifyText } from "@wesr/snip";

const result = classifyText(`from fastapi import FastAPI

app = FastAPI()

@app.get("/notes/{note_id}")
def read_note(note_id: int):
    return {"note_id": note_id}
`);
console.log(result.label);

Classifies a string using the embedded SNIP model.

interface ClassificationResult {
  label: SnipLabel;
  predicted_label: SnipLabel;
  confidence: number;
  margin: number;
  alternatives: Array<[SnipLabel, number]>;
}

• label is the accepted label after any configured fallback thresholds.
• predicted_label is the highest-scoring raw prediction.
• confidence is the score assigned to the winning label.
• margin is the gap between the winning label and the runner-up. Larger margins generally indicate a clearer classification.
• alternatives contains the top five labels and scores.

Yields once before classifying, which gives browser UI code an awaitable API and a chance to paint before SNIP runs.

import { classifyTextAsync } from "@wesr/snip";

const result = await classifyTextAsync("package main\n\nfunc main() {}\n");

Applies the model's bounded sampling strategy. Inputs up to 16 KiB are classified whole; larger inputs are represented by start, middle, and end windows.

The release model predicts:

bash, c, cpp, csharp, css, csv, diff, dockerfile, go, html, ini, java, javascript, json, log, lua, markdown, php, plain_text, powershell, python, ruby, rust, sql, toml, typescript, xml, yaml.

• Raw model JSON: 626,596 bytes
• Gzip model JSON: 203,820 bytes

The runtime is written in TypeScript and has no runtime package dependencies.

SNIP is released under the BSD 3-Clause License. See LICENSE.