WebGPU)¶

Models that run directly in the browser — no server, no Python, no setup. These are the models our tools can use.

Transformers.js Models¶

Transformers.js runs Hugging Face models in the browser via ONNX Runtime Web. All models below have ONNX exports.

Model	Task	Size	Speed	HF Link
distilbert-base-uncased	Sentiment/classification	67 MB	Fast	HF
bert-base-uncased	Fill-mask, embeddings	110 MB	Medium	HF
t5-small	Summarization, translation	60 MB	Medium	HF
all-MiniLM-L6-v2	Sentence embeddings	23 MB	Fast	HF
xenova/multilingual-e5-small	Multilingual embeddings	118 MB	Medium	HF
Xenova/distilgpt2	Text generation	88 MB	Medium	HF

Model	Task	Size	Speed	HF Link
mobilenet_v2	Image classification (1000 classes)	14 MB	Very fast	HF
detr-resnet-50	Object detection	160 MB	Slow	HF
vit-base-patch16-224	Image classification	330 MB	Slow	HF
clip-vit-base-patch32	Image-text similarity	150 MB	Medium	HF
sam-vit-base	Segment anything	380 MB	Slow	HF

Model	Task	Size	Speed	HF Link
whisper-tiny	Speech-to-text	39 MB	Fast	HF
whisper-small	Speech-to-text	244 MB	Medium	HF
wav2vec2-base	Speech recognition	95 MB	Medium	HF

Pre-trained models that run via TF.js in the browser. Good for transfer learning (fine-tuning).

Model	Task	Size	Transfer Learning?	Link
MobileNet v2	Image classification	7 MB	Yes — retrain last layer	tfhub.dev
COCO-SSD	Object detection	5 MB	No	github.com/tensorflow
BlazeFace	Face detection	400 KB	No	github.com/tensorflow
PoseNet	Body pose estimation	4-13 MB	No	github.com/tensorflow
Universal Sentence Encoder	Sentence embeddings	28 MB	No	tfhub.dev
Toxicity	Toxic text detection	25 MB	No	github.com/tensorflow

Optimized models in ONNX format, convertible to ONNX Runtime Web for browser inference.

Model	Task	Size	ONNX Zoo Link
ResNet-50	Image classification	97 MB	github.com/onnx/models
EfficientNet-Lite	Image classification	19 MB	github.com/onnx/models
SSD MobileNet	Object detection	27 MB	github.com/onnx/models
BERT-Squad	Question answering	130 MB	github.com/onnx/models
GPT-2	Text generation	600 MB	github.com/onnx/models

Models that leverage WebGPU for 10-50x faster inference than WebGL. Requires Chrome 113+ or Edge 113+.

Library	Models	WebGPU Support	Link
Transformers.js v3	All ONNX models above	Yes (via ONNX Runtime)	huggingface.co/docs/transformers.js
WebLLM	LLaMA, Mistral, Gemma, Phi	Yes (full LLM in browser)	webllm.mlc.ai
MediaPipe	Face, hand, pose, object	Yes	ai.google.dev/edge/mediapipe

Integration with FreeAgentStore

Many of these browser-ready models are already packaged as tools on FreeAgentStore. Check the Models tab for ready-to-use versions.