@xenova/transformers - npm Package Compare versions

Comparing version 1.0.3 to 1.0.4

package.json

 {
   "name": "@xenova/transformers",
-  "version": "1.0.3",
+  "version": "1.0.4",
   "description": "Run 🤗 Transformers in your browser! We currently support BERT, DistilBERT, T5, GPT2, and BART models, for a variety of tasks including: masked language modelling, text classification, translation, summarization, question answering, and text generation.",
@@ -5,0 +5,0 @@ "main": "./src/transformers.js",

README.md

@@ -7,6 +7,42 @@ # Transformers.js
 
+Check out our demo at [https://xenova.github.io/transformers.js/](https://xenova.github.io/transformers.js/). As you'll see, everything runs inside the browser!
 
-
 ## Getting Started
 
+### Installation
+If you use [npm](https://www.npmjs.com/package/@xenova/transformers), you can install it using:
+```bash
+npm i @xenova/transformers
+```
+
+Alternatively, you can use it in a `<script>` tag from a CDN, for example:
+```html
+<!-- Using jsDelivr -->
+<script src="https://cdn.jsdelivr.net/npm/@xenova/transformers/dist/transformers.min.js"></script>
+
+<!-- or UNPKG -->
+<script src="https://www.unpkg.com/@xenova/transformers/dist/transformers.min.js"></script>
+```
+
+### Setup
+By default, Transformers.js uses hosted [models](https://huggingface.co/Xenova/transformers.js/tree/main/quantized) precompiled [WASM binaries](https://cdn.jsdelivr.net/npm/@xenova/transformers/dist/), which should work out-of-the-box. You can override this behaviour as follows:
+```javascript
+import { env } from "@xenova/transformers";
+
+// Use a different host for models.
+// - `remoteURL` defaults to use the HuggingFace Hub
+// - `localURL` defaults to '/models/onnx/quantized/'
+env.remoteURL = 'https://www.example.com/';
+env.localURL = '/path/to/models/';
+
+// Set whether to use remote or local models. Defaults to true.
+//  - If true, use the path specified by `env.remoteURL`.
+//  - If false, use the path specified by `env.localURL`.
+env.remoteModels = false;
+
+// Set parent path of .wasm files. Defaults to use a CDN.
+env.onnx.wasm.wasmPaths = '/path/to/files/';
+```
+
+### Quick tour
 It's super easy to translate from existing code!
@@ -37,19 +73,8 @@
 
-*Note:* If running locally, it is assumed that the required model files are located in `./models/onnx/quantized/`. To override this behaviour, you can specify the model path or URL as a second argument to the pipeline function. For example, to use models from the HuggingFace hub:
-
+*Note:* In the same way as the Python library, you can use a different model by providing its name as the second argument to the pipeline function. For example:
 ```javascript
-// Set host, model_id and task:
-const hf_url = 'https://huggingface.co/Xenova/transformers.js/resolve/main/quantized';
-const model_id = 'distilbert-base-uncased-finetuned-sst-2-english';
-const task = 'sequence-classification';
-
-const model_url = `${hf_url}/${model_id}/${task}`;
-
-// You can now create the classifier using:
-let classifier = await pipeline('sentiment-analysis', model_url);
+// Use a different model for sentiment-analysis
+let classifier = await pipeline('sentiment-analysis', 'nlptown/bert-base-multilingual-uncased-sentiment');
 ```
 
-## Demo
-Check out our demo at [https://xenova.github.io/transformers.js/](https://xenova.github.io/transformers.js/). As you'll see, everything runs inside the browser!
-
 ## Usage
@@ -56,0 +81,0 @@

src/pipelines.js

@@ -5,3 +5,4 @@ const {
     getTopItems,
-    cos_sim
+    cos_sim,
+    pathJoin
 } = require("./utils.js");
@@ -21,5 +22,6 @@
 
+const {
+    env
+} = require('./env.js')
 
-const HF_HUB_MODEL_PATH_TEMPLATE = 'https://huggingface.co/Xenova/transformers.js/resolve/main/quantized/{model}/{task}/';
-const DEFAULT_MODEL_PATH_TEMPLATE = '/models/onnx/quantized/{model}/{task}';
 
@@ -329,7 +331,17 @@ class Pipeline extends Callable {
 const TASK_NAME_MAPPING = {
-    // Fix mismatch between pipeline name and exports (folder name)
+    // Fix mismatch between pipeline's task name and exports (folder name)
     'text-classification': 'sequence-classification',
-    'embeddings': 'default'
+    'embeddings': 'default',
+    'fill-mask': 'masked-lm',
+
+    'summarization': 'seq2seq-lm-with-past',
+    'text-generation': 'causal-lm-with-past',
 }
 
+const TASK_PREFIX_MAPPING = {
+    // if task starts with one of these, set the corresponding folder name
+    'translation': 'seq2seq-lm-with-past',
+}
+
+
 const TASK_ALIASES = {
@@ -346,3 +358,2 @@ "sentiment-analysis": "text-classification",
     {
-        default_model_path_template = DEFAULT_MODEL_PATH_TEMPLATE,
         progress_callback = null
@@ -362,12 +373,33 @@ } = {}
 
+
     // Use model if specified, otherwise, use default
-    let modelPath = model;
-    if (!modelPath) {
-        modelPath = default_model_path_template
-            .replace('{model}', pipelineInfo.default.model)
-            .replace('{task}', TASK_NAME_MAPPING[task] ?? task);
+    if (!model) {
+        model = pipelineInfo.default.model
+        console.log(`No model specified. Using default model: "${model}".`);
+    }
 
-        console.log(`No model specified. Attempting to load default model from ${default_model_path_template}.`);
+    // determine suffix
+    let suffix = TASK_NAME_MAPPING[task];
+    if (!suffix) {
+        // try get from suffix
+        for (const [prefix, mapping] of Object.entries(TASK_PREFIX_MAPPING)) {
+            if (task.startsWith(prefix)) {
+                suffix = mapping;
+                break;
+            }
+        }
     }
 
+    if (!suffix) {
+        // Still not set... so, we default to the name given
+        suffix = task;
+    }
+
+    // Construct model path
+    model = pathJoin(
+        (env.remoteModels) ? env.remoteURL : env.localURL, // host prefix
+        model, // model name
+        suffix, // task suffix
+    )
+
     let modelClass = pipelineInfo.model;
@@ -378,4 +410,4 @@ let pipelineClass = pipelineInfo.pipeline;
     let [pipelineTokenizer, pipelineModel] = await Promise.all([
-        AutoTokenizer.from_pretrained(modelPath, progress_callback),
-        modelClass.from_pretrained(modelPath, progress_callback)
+        AutoTokenizer.from_pretrained(model, progress_callback),
+        modelClass.from_pretrained(model, progress_callback)
     ])
@@ -382,0 +414,0 @@

src/transformers.js

@@ -21,4 +21,5 @@
 } = require("./pipelines.js");
-const { env } = require('onnxruntime-web');
+const { env } = require('./env.js');
 
+
 const moduleExports = {
@@ -44,3 +45,3 @@ // Tokenizers
 
-    // onnx runtime web env
+    // environment variables
     env
@@ -47,0 +48,0 @@ };

src/utils.js

 
 const fs = require('fs');
 
-// Use caching when available
-const CACHE_AVAILABLE = typeof self !== 'undefined' && 'caches' in self;
-const FS_AVAILABLE = !isEmpty(fs); // check if file system is available
+const { env } = require('./env.js');
 
@@ -118,3 +116,3 @@ class FileResponse {
 
-    if (FS_AVAILABLE && !isValidHttpUrl(url)) {
+    if (env.useFS && !isValidHttpUrl(url)) {
         return new FileResponse(url)
@@ -131,6 +129,2 @@
 
-function isEmpty(obj) {
-    return Object.keys(obj).length === 0;
-}
-
 async function getModelFile(modelPath, fileName, progressCallback = null) {
@@ -146,3 +140,3 @@
     let cache;
-    if (CACHE_AVAILABLE) {
+    if (env.useCache) {
         cache = await caches.open('transformers-cache');
@@ -156,3 +150,3 @@ }
 
-    if (!CACHE_AVAILABLE || (response = await cache.match(request)) === undefined) {
+    if (!env.useCache || (response = await cache.match(request)) === undefined) {
         // Caching not available, or model is not cached, so we perform the request
@@ -165,3 +159,3 @@ response = await getFile(request);
 
-        if (CACHE_AVAILABLE) {
+        if (env.useCache) {
             // only clone if cache available
@@ -168,0 +162,0 @@ responseToCache = response.clone();

tests/index.js

@@ -6,3 +6,2 @@
 
-const path = require('path')
 const { pipeline, env } = require('..')
@@ -12,11 +11,8 @@
 // This is done by setting numThreads to 1
-env.wasm.numThreads = 1
+env.onnx.wasm.numThreads = 1
 
-// Set base model dir for testing.
-const BASE_MODEL_DIR = path.join(path.dirname(__dirname), '/models/onnx/quantized/')
+// Only use local models
+env.remoteModels = false;
+// env.remoteModels = true; // Uncomment to test online
 
-// Uncomment to test online
-// const BASE_MODEL_DIR = 'https://huggingface.co/Xenova/transformers.js/resolve/main/quantized/'
-
-
 function isDeepEqual(obj1, obj2, {
@@ -60,6 +56,5 @@ tol = 1e-3
 async function embeddings() {
-    let model_path = path.join(BASE_MODEL_DIR, 'sentence-transformers/all-MiniLM-L6-v2/default')
 
     // Load embeddings pipeline (uses  by default)
-    let embedder = await pipeline('embeddings', model_path)
+    let embedder = await pipeline('embeddings', 'sentence-transformers/all-MiniLM-L6-v2')
 
@@ -94,6 +89,4 @@
 async function text_classification() {
-    let model_path = path.join(BASE_MODEL_DIR, 'distilbert-base-uncased-finetuned-sst-2-english/sequence-classification')
+    let classifier = await pipeline('text-classification', 'distilbert-base-uncased-finetuned-sst-2-english');
 
-    let classifier = await pipeline('text-classification', model_path);
-
     let outputs1 = await classifier("I hated the movie");
@@ -144,5 +137,4 @@
 async function masked_language_modelling() {
-    let model_path = path.join(BASE_MODEL_DIR, 'bert-base-uncased/masked-lm')
 
-    let unmasker = await pipeline('fill-mask', model_path);
+    let unmasker = await pipeline('fill-mask', 'bert-base-uncased');
 
@@ -184,3 +176,2 @@ let outputs1 = await unmasker("Once upon a [MASK].");
 async function question_answering() {
-    let model_path = path.join(BASE_MODEL_DIR, 'distilbert-base-uncased-distilled-squad/question-answering')
 
@@ -190,3 +181,3 @@ let question = 'Who was Jim Henson?'
 
-    let answerer = await pipeline('question-answering', model_path);
+    let answerer = await pipeline('question-answering', 'distilbert-base-uncased-distilled-squad');
     let outputs = await answerer(question, context);
@@ -212,5 +203,4 @@
 async function summarization() {
-    let model_path = path.join(BASE_MODEL_DIR, 'sshleifer/distilbart-cnn-6-6/seq2seq-lm-with-past')
 
-    let summarizer = await pipeline('summarization', model_path)
+    let summarizer = await pipeline('summarization', 'sshleifer/distilbart-cnn-6-6')
 
@@ -233,5 +223,4 @@ let texts = [
 async function translation() {
-    let model_path = path.join(BASE_MODEL_DIR, 't5-small/seq2seq-lm-with-past')
 
-    let translator = await pipeline('translation_en_to_de', model_path)
+    let translator = await pipeline('translation_en_to_de', 't5-small')
 
@@ -260,5 +249,4 @@ let translation1 = await translator('Hello, how are you?')
 async function text_generation() {
-    let model_path = path.join(BASE_MODEL_DIR, 'distilgpt2/causal-lm-with-past')
 
-    let generator = await pipeline('text-generation', model_path)
+    let generator = await pipeline('text-generation', 'distilgpt2')
 
@@ -317,2 +305,2 @@ let output1 = await generator('Once upon a time, there was a', {
     console.log('Embeddings:', await embeddings())
-})();
+})();

New alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Fixed alerts

License Policy Violation

License

This package is not allowed per your license policy. Review the package's license to ensure compliance.

Found 1 instance in 1 package

Improved metrics

Total package byte prevSize

41909071

increased by0.03%

Number of package files

19

increased by5.56%

Lines of code

10676

increased by1.25%

Number of lines in readme file

110

increased by29.41%

Worsened metrics

Number of low supply chain risk alerts

2

increased by100%

No dependency changes