Merge pull request #10 from jgw96/better-tests

v0.2.1

Author: jgw96

README.md

@@ -13,17 +13,20 @@ npm install web-ai-toolkit
 
 ## Available Functions
 
+*Note: Supported hardware is listed in priority of device selection. For example, for transcribing an audio file,
+the code will attempt to choose an NPU first, then a GPU and finally the CPU if neither of the first two are found.*
+
 | Function Name         | Parameter      | Type                   | Default Value | Supported Hardware |
 |-----------------------|----------------|------------------------|---------------|--------------------|
-| transcribeAudioFile   | audioFile      | Blob                   | -             | NPU                |
+| transcribeAudioFile   | audioFile      | Blob                   | -             | NPU / GPU / CPU               |
 |                       | model          | string                 | "Xenova/whisper-tiny"|                    |
 |                       | timestamps     | boolean                | false         |                    |
 |                       | language       | string                 | "en-US"       |                    |
-| textToSpeech          | text           | string                 | -             | GPU                |
+| textToSpeech          | text           | string                 | -             | GPU / CPU               |
 |                       | model          | string                 | "Xenova/mms-tts-eng"|                    |
-| summarize             | text           | string                 | -             | GPU                |
+| summarize             | text           | string                 | -             | GPU / CPU               |
 |                       | model          | string                 | "Xenova/distilbart-cnn-6-6"|                |
-| ocr                   | image          | Blob                   | -             | GPU                |
+| ocr                   | image          | Blob                   | -             | GPU / CPU               |
 |                       | model          | string                 | "Xenova/trocr-small-printed"|                 |
 
 ## Technical Details

package.json

@@ -1,6 +1,6 @@
 {
   "name": "web-ai-toolkit",
-  "version": "0.2.0",
+  "version": "0.2.1",
   "repository": "https://github.com/jgw96/web-ai-toolkit",
   "keywords": [
     "ai",

src/services/ocr/ocr.ts

@@ -1,5 +1,6 @@
 /* eslint-disable no-async-promise-executor */
 import { pipeline, env } from '@huggingface/transformers';
+import { webGPUCheck } from '../../utils';
 
 let ocr: any = undefined;
 
@@ -29,7 +30,7 @@ async function loadOCR(model: string): Promise<void> {
             env.allowLocalModels = false;
             env.useBrowserCache = false;
             ocr = await pipeline('image-to-text', model || 'Xenova/trocr-small-printed', {
-                device: (navigator as any).ml ? "webnn" : "webgpu"
+                device: (navigator as any).ml ? "webnn" : await webGPUCheck() ? "webgpu" : "wasm"
             });
             console.log("loaded ocr", ocr)
             resolve();

src/services/speech-recognition/recognition.ts

@@ -1,5 +1,6 @@
 /* eslint-disable no-async-promise-executor */
 import { AutomaticSpeechRecognitionPipeline, pipeline, env } from '@huggingface/transformers';
+import { webGPUCheck } from '../../utils';
 
 let transcriber: AutomaticSpeechRecognitionPipeline | undefined = undefined;
 
@@ -57,7 +58,7 @@ export async function loadTranscriber(model: string = "Xenova/whisper-tiny", tim
                 // @ts-ignore
                 return_timestamps: timestamps,
                 language,
-                device: (navigator as any).ml ? "webnn" : "webgpu"
+                device: (navigator as any).ml ? "webnn" : await webGPUCheck() ? "webgpu" : "wasm"
             });
 
             resolve();

src/services/summarization/summarization.ts

@@ -1,4 +1,5 @@
 import { pipeline, env } from '@huggingface/transformers';
+import { webGPUCheck } from '../../utils';
 
 let summarizer: any = undefined;
 
@@ -26,7 +27,7 @@ async function loadSummarizer(model: string): Promise<void> {
 
             summarizer = await pipeline('summarization', model || 'Xenova/distilbart-cnn-6-6', {
                 dtype: "fp32",
-                device: (navigator as any).ml ? "webnn" : "webgpu"
+                device: (navigator as any).ml ? "webnn" : await webGPUCheck() ? "webgpu" : "wasm"
             });
             console.log("loaded summarizer", summarizer)
             resolve();

src/utils.ts

@@ -0,0 +1,16 @@
+export async function webGPUCheck(): Promise<boolean> {
+    // check to see if navigator.gpu exists and if we can create a device
+    if ((navigator as any).gpu) {
+        try {
+            const gpu = await (navigator as any).gpu.requestAdapter();
+            return gpu ? true : false;
+        }
+        catch (err) {
+            console.error(err);
+            return false;
+        }
+    }
+    else {
+        return false;
+    }
+}