[WIP] Transformers.js improvements

- default to q4 since the model is quite large
- support external data format (see [here](https://github.com/huggingface/transformers.js/pull/1212))

config.json

@@ -25,9 +25,14 @@
   "torch_dtype": "bfloat16",
   "transformers_version": "4.42.3",
   "transformers.js_config": {
+    "dtype": "q4",
     "kv_cache_dtype": {
       "q4f16": "float16",
       "fp16": "float16"
+    },
+    "use_external_data_format": {
+      "model.onnx": true,
+      "model_fp16.onnx": true
     }
   },
   "use_cache": true,