Transformers.js v3.5 optimizations (#2)

- Transformers.js v3.5 optimizations (9c78ee8eec92f0cbabdffcbd91d62f4853d4e806)
- Update config.json (b766e9e06f6a8777736101c0c91b1def50badc10)

README.md

@@ -12,9 +12,9 @@ https://github.com/apple/ml-mobileclip with ONNX weights to be compatible with T
 
 ## Usage (Transformers.js)
 
-If you haven't already, you can install the [Transformers.js](https://huggingface.co/docs/transformers.js) JavaScript library from [NPM](https://www.npmjs.com/package/@xenova/transformers) using:
+If you haven't already, you can install the [Transformers.js](https://huggingface.co/docs/transformers.js) JavaScript library from [NPM](https://www.npmjs.com/package/@huggingface/transformers) using:
 ```bash
-npm i @xenova/transformers
+npm i @huggingface/transformers
 ```
 
 **Example:** Perform zero-shot image classification.
@@ -27,7 +27,7 @@ import {
   RawImage,
   dot,
   softmax,
-} from '@xenova/transformers';
+} from '@huggingface/transformers';
 
 const model_id = 'Xenova/mobileclip_blt';
 
@@ -37,9 +37,7 @@ const text_model = await CLIPTextModelWithProjection.from_pretrained(model_id);
 
 // Load processor and vision model
 const processor = await AutoProcessor.from_pretrained(model_id);
-const vision_model = await CLIPVisionModelWithProjection.from_pretrained(model_id, {
-  quantized: false, // NOTE: vision model is sensitive to quantization.
-});
+const vision_model = await CLIPVisionModelWithProjection.from_pretrained(model_id);
 
 // Run tokenization
 const texts = ['cats', 'dogs', 'birds'];

config.json

@@ -1,3 +1,15 @@
 {
-  "model_type": "clip"
+  "model_type": "clip",
+  "transformers.js_config": {
+    "dtype": {
+      "vision_model": "fp32"
+    },
+    "device_config": {
+      "webnn": {
+        "free_dimension_overrides": {
+          "batch_size": 1
+        }
+      }
+    }
+  }
 }