Runs in the browser · Vision
Run SigLIP2 Base Patch16-224 in your browser
SigLIP2 Base Patch16-224 downloads 948.7 MB over WebGPU (q4f16), or 721.0 MB over WebAssembly (uint8), to classify images against your own text labels entirely in the tab with Transformers.js. No install, no server.
Vision · onnx-community/siglip2-base-patch16-224-ONNX
- Parameters
- 375.19M
- Pipeline task
- zero-shot-image-classification
Reading
The WASM build downloads smaller here: 721.0 MB (uint8) against 948.7 MB (q4f16) for WebGPU. The CPU-fallback quant compresses tighter than the GPU pick for this model.
The largest vision download in the catalog (948.7 MB).
2.53 MB/M params against a 1.73 MB/M vision median: dense for its parameter count.
The quant ladder spans 4.0x: 721.0 MB (uint8) to 2.80 GB (fp32).
Will it run in your browser?
Checking for WebGPU support…
All measured variants
| Variant | Size |
|---|---|
| uint8 WASM pick | 721.0 MB |
| q4f16 WebGPU pick | 948.7 MB |
| fp16 | 1.40 GB |
| q8 | 1.41 GB |
| bnb4 | 1.67 GB |
| q4 | 1.69 GB |
| fp32 | 2.80 GB |
Sizes measured from the HuggingFace API file tree for onnx-community/siglip2-base-patch16-224-ONNX, not estimated.
Use it with Transformers.js
import { pipeline } from "@huggingface/transformers";
const pipe = await pipeline("zero-shot-image-classification", "onnx-community/siglip2-base-patch16-224-ONNX", {
device: "webgpu", // usually falls back to "wasm"; wrap in try/catch for production
dtype: "q4f16", // use "uint8" for the WASM build
});
const result = await pipe(/* your input */);
Requires npm install @huggingface/transformers (or the CDN build).
Sources
Same job, different size
FAQ
WebGPU or WASM for SigLIP2 Base Patch16-224, 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 (uint8) on CPU: same model, slower to load, slower to run.
What does q4f16 mean for SigLIP2 Base Patch16-224?
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 uint8: 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.