Runs in the browser · Multimodal (any-to-any)
Run Gemma-4-E2B-it in your browser
Gemma-4-E2B-it downloads 3.15 GB over WebGPU (q4f16), or 2.89 GB over WebAssembly (q8), to handle text, image, and audio from one checkpoint entirely in the tab with Transformers.js. No install, no server.
Multimodal (any-to-any) · onnx-community/gemma-4-E2B-it-ONNX
- Parameters
- 5.12B
- Pipeline task
- custom
Reading
The E2B name is an effective/active-param label (like Gemma 3n), not the real stored parameter count: this checkpoint stores about 5.12B params, not 2B. Any-to-any multimodal generation (text, image and audio in and out) has no stable transformers.js pipeline() task yet; expect to wire the encoders and decoder manually. This is a multi-gigabyte download. Between the file size and current browser memory limits, it is impractical to run on most machines today.
The WASM build downloads smaller here: 2.89 GB (q8) against 3.15 GB (q4f16) for WebGPU. The CPU-fallback quant compresses tighter than the GPU pick for this model.
Will it run in your browser?
Checking for WebGPU support…
All measured variants
| Variant | Size |
|---|---|
| q8 WASM pick | 2.89 GB |
| q4f16 WebGPU pick | 3.15 GB |
| fp16 | 3.56 GB |
| q4 | 3.66 GB |
| fp32 | 5.15 GB |
Sizes measured from the HuggingFace API file tree for onnx-community/gemma-4-E2B-it-ONNX, not estimated.
Use it with Transformers.js
There is no stable transformers.js pipeline() call for this model yet, and no verified low-level snippet either (see the Reading panel above). Check the model card on HuggingFace for current usage examples.
Sources
Same job, different size
FAQ
WebGPU or WASM for Gemma-4-E2B-it, in practice?
Chrome, Edge, and Safari 26+ run the WebGPU build (q4f16) on the GPU. Firefox has WebGPU on Windows and Apple Silicon Macs but not everywhere yet. Any browser without a WebGPU adapter falls back automatically to the WebAssembly build (q8) on CPU: same model, slower to load, slower to run.
What does q4f16 mean for Gemma-4-E2B-it?
The WebGPU build here uses q4f16: 4-bit weights with some layers kept at 16-bit for stability: WebGPU's usual smallest clean build. The WASM fallback uses q8: 8-bit integers: about a quarter the size of fp32, with a smaller quality trade than 4-bit.
Sizes measured 2026-08-01 from the HuggingFace API. Last validated 2026-08-03. See all browser models or the methodology.