Added knip to the test pipeline

.github/workflows/tests.yml

@@ -18,6 +18,23 @@ env:
   ONNXRUNTIME_NODE_INSTALL: skip
 
 jobs:
+  knip:
+    if: github.event.pull_request.draft == false
+    runs-on: ubuntu-latest
+
+    steps:
+      - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd  # v6.0.2
+      - name: Install pnpm
+        uses: pnpm/action-setup@b906affcce14559ad1aafd4ab0e942779e9f58b1  # v4
+      - name: Use Node.js 24.10.0
+        uses: actions/setup-node@53b83947a5a98c8d113130e565377fae1a50d02f  # v6
+        with:
+          node-version: "24.10.0"
+          cache: "pnpm"
+      - run: pnpm install --frozen-lockfile
+      - run: pnpm format:check
+      - run: pnpm knip
+
   build:
     if: github.event.pull_request.draft == false
     runs-on:

package.json

@@ -10,7 +10,8 @@
     "test": "pnpm -r test",
     "format": "prettier --write .",
     "format:check": "prettier --check .",
-    "dev": "node scripts/dev.mjs"
+    "dev": "node scripts/dev.mjs",
+    "knip": "knip"
   },
   "repository": {
     "type": "git",
@@ -23,9 +24,49 @@
   },
   "homepage": "https://github.com/huggingface/transformers.js#readme",
   "devDependencies": {
+    "@types/node": "^24.10.9",
+    "knip": "6.15.0",
     "prettier": "3.8.1",
     "typescript": "5.9.3"
   },
+  "knip": {
+    "tags": [
+      "-lintignore"
+    ],
+    "ignoreBinaries": [
+      "doc-builder",
+      "python"
+    ],
+    "workspaces": {
+      ".": {
+        "entry": [
+          "scripts/*.mjs"
+        ],
+        "project": [
+          "scripts/**/*.mjs"
+        ]
+      },
+      "packages/transformers": {
+        "entry": [
+          "src/models/registry.js",
+          "tests/**/*.js",
+          "scripts/*.mjs",
+          "docs/scripts/**/*.js",
+          "docs/plugins/**/*.js"
+        ],
+        "project": [
+          "src/**/*.{js,ts}",
+          "tests/**/*.js",
+          "scripts/**/*.mjs",
+          "docs/scripts/**/*.js",
+          "docs/plugins/**/*.js",
+          "!coverage/**",
+          "!dist/**",
+          "!types/**"
+        ]
+      }
+    }
+  },
   "prettier": {
     "overrides": [
       {

packages/transformers/package.json

@@ -57,17 +57,18 @@
   "dependencies": {
     "@huggingface/jinja": "^0.5.6",
     "@huggingface/tokenizers": "^0.1.3",
+    "onnxruntime-common": "1.24.3",
     "onnxruntime-node": "1.24.3",
     "onnxruntime-web": "1.26.0-dev.20260416-b7804b056c",
     "sharp": "^0.34.5"
   },
   "devDependencies": {
+    "@jest/globals": "30.2.0",
     "@types/jest": "^30.0.0",
     "@types/node": "^24.1.0",
     "@webgpu/types": "^0.1.69",
     "esbuild": "^0.27.2",
     "jest": "^30.2.0",
-    "jest-environment-node": "^30.2.0",
     "jsdoc-to-markdown": "^9.1.3",
     "typescript": "5.9.3"
   },

packages/transformers/scripts/build/constants.mjs

@@ -1,7 +1,7 @@
 import path from "node:path";
 import { fileURLToPath } from "node:url";
 
-export const DIST_FOLDER = "dist";
+const DIST_FOLDER = "dist";
 export const NODE_IGNORE_MODULES = ["onnxruntime-web"];
 export const NODE_EXTERNAL_MODULES = [
   "onnxruntime-common",

packages/transformers/src/models/modeling_utils.js

@@ -86,7 +86,7 @@ export function boolTensor(value) {
     return new Tensor('bool', [value], [1]);
 }
 
-export { getSessionsConfig, getTextOnlySessions, MODEL_TYPES } from './session_config.js';
+export { MODEL_TYPES } from './session_config.js';
 
 /**
  * Runtime-only model type configuration (forward functions, generation flags).
@@ -1101,7 +1101,7 @@ export class PreTrainedModel extends Callable {
  * @returns {Promise<Seq2SeqLMOutput>} Promise that resolves with the output of the seq2seq model.
  * @private
  */
-export async function seq2seq_forward(self, model_inputs) {
+async function seq2seq_forward(self, model_inputs) {
     let { encoder_outputs, input_ids, decoder_input_ids, decoder_attention_mask, ...other_decoder_inputs } =
         model_inputs;
     // Encode if needed
@@ -1164,7 +1164,7 @@ export async function encoder_forward(self, model_inputs) {
     return await sessionRun(session, encoderFeeds);
 }
 
-export async function auto_encoder_forward(self, model_inputs) {
+async function auto_encoder_forward(self, model_inputs) {
     const encoded = await self.encode(model_inputs);
     const decoded = await self.decode(encoded);
     return decoded;
@@ -1219,7 +1219,7 @@ export function getPastKeyValues(decoderResults, pastKeyValues) {
  * @param {Object} model_output The output of the model.
  * @returns {{cross_attentions?: Tensor[]}} An object containing attentions.
  */
-export function getAttentions(model_output) {
+function getAttentions(model_output) {
     const attentions = {};
 
     for (const attnName of ['cross_attentions', 'encoder_attentions', 'decoder_attentions']) {
@@ -1376,7 +1376,7 @@ export async function decoder_forward(self, model_inputs, is_encoder_decoder = f
  * @returns {Promise<Tensor>} The model's output tensor
  * @private
  */
-export async function generic_text_to_text_forward(
+async function generic_text_to_text_forward(
     self,
     {
         // Generic parameters:
@@ -1489,7 +1489,7 @@ export async function generic_text_to_text_forward(
  * @returns {Promise<Tensor>} The model's output tensor.
  * @private
  */
-export async function audio_text_to_text_forward(self, params) {
+async function audio_text_to_text_forward(self, params) {
     return await generic_text_to_text_forward(self, {
         ...params,
         modality_input_names: ['audio_values', 'input_features'],
@@ -1506,7 +1506,7 @@ export async function audio_text_to_text_forward(self, params) {
  * @returns {Promise<Tensor>} The model's output tensor.
  * @private
  */
-export async function image_text_to_text_forward(self, params) {
+async function image_text_to_text_forward(self, params) {
     return await generic_text_to_text_forward(self, {
         ...params,
         modality_input_names: ['pixel_values'],
@@ -1559,7 +1559,7 @@ export function cumsum_masked_fill(attention_mask, start_index = 0) {
  *     position_ids = position_ids[:, -input_ids.shape[1] :]
  * ```
  */
-export function create_position_ids(model_inputs, past_key_values = null, start_index = 0) {
+function create_position_ids(model_inputs, past_key_values = null, start_index = 0) {
     const { input_ids, inputs_embeds, attention_mask } = model_inputs;
 
     const { data, dims } = cumsum_masked_fill(attention_mask, start_index);
@@ -1631,15 +1631,15 @@ export function encoder_decoder_prepare_inputs_for_generation(self, input_ids, m
     };
 }
 
-export function multimodal_text_to_text_prepare_inputs_for_generation(self, ...args) {
+function multimodal_text_to_text_prepare_inputs_for_generation(self, ...args) {
     if (self.config.is_encoder_decoder) {
         return encoder_decoder_prepare_inputs_for_generation(self, ...args);
     } else {
         return decoder_prepare_inputs_for_generation(self, ...args);
     }
 }
 
-export function default_merge_input_ids_with_features({
+function default_merge_input_ids_with_features({
     modality_token_id,
     inputs_embeds,
     modality_features,
@@ -1714,7 +1714,7 @@ export function default_merge_input_ids_with_audio_features({
  * @returns {Promise<Record<string, any>>} A Promise that resolves to a dictionary of configuration objects.
  * @private
  */
-export async function get_optional_configs(pretrained_model_name_or_path, names, options) {
+async function get_optional_configs(pretrained_model_name_or_path, names, options) {
     return Object.fromEntries(
         await Promise.all(
             Object.keys(names).map(async (name) => {

packages/transformers/src/models/whisper/common_whisper.js

@@ -103,7 +103,7 @@ const WHISPER_LANGUAGES = [
 // @ts-ignore
 export const WHISPER_LANGUAGE_MAPPING = new Map(WHISPER_LANGUAGES);
 // @ts-ignore
-export const WHISPER_TO_LANGUAGE_CODE_MAPPING = new Map([
+const WHISPER_TO_LANGUAGE_CODE_MAPPING = new Map([
     ...WHISPER_LANGUAGES.map(([k, v]) => [v, k]),
     ...[
         ['burmese', 'my'],

packages/transformers/src/utils/audio.js

@@ -77,7 +77,9 @@ export async function load_audio(url, sampling_rate) {
 /**
  * @deprecated Use {@link load_audio} instead.
  */
-export const read_audio = load_audio;
+export async function read_audio(url, sampling_rate) {
+    return await load_audio(url, sampling_rate);
+}
 
 /**
  * Helper function to generate windows that are special cases of the generalized cosine window.

packages/transformers/src/utils/constants.js

@@ -1,8 +1,6 @@
 export const GITHUB_ISSUE_URL = 'https://github.com/huggingface/transformers.js/issues/new/choose';
 
-export const CONFIG_NAME = 'config.json';
 export const FEATURE_EXTRACTOR_NAME = 'preprocessor_config.json';
-export const IMAGE_PROCESSOR_NAME = FEATURE_EXTRACTOR_NAME;
+export const IMAGE_PROCESSOR_NAME = 'preprocessor_config.json';
 export const PROCESSOR_NAME = 'processor_config.json';
 export const CHAT_TEMPLATE_NAME = 'chat_template.jinja';
-export const GENERATION_CONFIG_NAME = 'generation_config.json';

packages/transformers/src/utils/core.js

@@ -141,39 +141,6 @@ export class DefaultProgressCallback extends Callable {
     }
 }
 
-/**
- * Reverses the keys and values of an object.
- *
- * @param {Object} data The object to reverse.
- * @returns {Object} The reversed object.
- * @see https://ultimatecourses.com/blog/reverse-object-keys-and-values-in-javascript
- */
-export function reverseDictionary(data) {
-    // https://ultimatecourses.com/blog/reverse-object-keys-and-values-in-javascript
-    return Object.fromEntries(Object.entries(data).map(([key, value]) => [value, key]));
-}
-
-/**
- * Escapes regular expression special characters from a string by replacing them with their escaped counterparts.
- *
- * @param {string} string The string to escape.
- * @returns {string} The escaped string.
- */
-export function escapeRegExp(string) {
-    return string.replace(/[.*+?^${}()|[\]\\]/g, '\\$&'); // $& means the whole matched string
-}
-
-/**
- * Check if a value is a typed array.
- * @param {*} val The value to check.
- * @returns {boolean} True if the value is a `TypedArray`, false otherwise.
- *
- * Adapted from https://stackoverflow.com/a/71091338/13989043
- */
-export function isTypedArray(val) {
-    return val?.prototype?.__proto__?.constructor?.name === 'TypedArray';
-}
-
 /**
  * Check if a value is an integer.
  * @param {*} x The value to check.
@@ -208,26 +175,6 @@ export function calculateDimensions(arr) {
     return dimensions;
 }
 
-/**
- * Replicate python's .pop() method for objects.
- * @param {Object} obj The object to pop from.
- * @param {string} key The key to pop.
- * @param {*} defaultValue The default value to return if the key does not exist.
- * @returns {*} The value of the popped key.
- * @throws {Error} If the key does not exist and no default value is provided.
- */
-export function pop(obj, key, defaultValue = undefined) {
-    const value = obj[key];
-    if (value !== undefined) {
-        delete obj[key];
-        return value;
-    }
-    if (defaultValue === undefined) {
-        throw Error(`Key ${key} does not exist in object.`);
-    }
-    return defaultValue;
-}
-
 /**
  * Efficiently merge arrays, creating a new copy.
  * Adapted from https://stackoverflow.com/a/6768642/13989043
@@ -277,18 +224,6 @@ export function pick(o, props) {
     );
 }
 
-/**
- * Calculate the length of a string, taking multi-byte characters into account.
- * This mimics the behavior of Python's `len` function.
- * @param {string} s The string to calculate the length of.
- * @returns {number} The length of the string.
- */
-export function len(s) {
-    let length = 0;
-    for (const c of s) ++length;
-    return length;
-}
-
 /**
  * Count the occurrences of a value in an array or string.
  * This mimics the behavior of Python's `count` method.

packages/transformers/src/utils/dtypes.js

@@ -49,8 +49,8 @@ export const DATA_TYPES = Object.freeze({
 });
 /** @typedef {keyof typeof DATA_TYPES} DataType */
 
-export const DEFAULT_DEVICE_DTYPE = DATA_TYPES.fp32;
-export const DEFAULT_DEVICE_DTYPE_MAPPING = Object.freeze({
+const DEFAULT_DEVICE_DTYPE = DATA_TYPES.fp32;
+const DEFAULT_DEVICE_DTYPE_MAPPING = Object.freeze({
     // NOTE: If not specified, will default to fp32
     [DEVICE_TYPES.wasm]: DATA_TYPES.q8,
 });

packages/transformers/src/utils/hub.js

@@ -195,7 +195,7 @@ export async function checkCachedResource(cache, localPath, proposedCacheKey) {
  * @param {PretrainedOptions} [options] Options containing progress callback and context for progress updates.
  * @returns {Promise<void>}
  */
-export async function storeCachedResource(path_or_repo_id, filename, cache, cacheKey, response, result, options = {}) {
+async function storeCachedResource(path_or_repo_id, filename, cache, cacheKey, response, result, options = {}) {
     // Check again whether request is in cache. If not, we add the response to the cache
     if ((await cache.match(cacheKey)) !== undefined) {
         return;
@@ -249,7 +249,7 @@ export async function storeCachedResource(path_or_repo_id, filename, cache, cach
  * @throws Will throw an error if the file is not found and `fatal` is true.
  * @returns {Promise<string|Uint8Array|null>} A Promise that resolves with the file content as a Uint8Array if `return_path` is false, or the file path as a string if `return_path` is true.
  */
-export async function loadResourceFile(
+async function loadResourceFile(
     path_or_repo_id,
     filename,
     fatal = true,

packages/transformers/tests/tokenizers.test.js

@@ -8,7 +8,7 @@ describe("Tokenizers (model-specific)", () => {
     describe(tokenizer_name, () => {
       for (const model_id in TEST_CONFIG) {
         describe(model_id, () => {
-          /** @type {import('../src/tokenizers.js').PreTrainedTokenizer} */
+          /** @type {import('../src/tokenization_utils.js').PreTrainedTokenizer} */
           let tokenizer;
           beforeAll(async () => {
             tokenizer = await TOKENIZER_CLASS.from_pretrained(model_id);