Runs in the browser · Embeddings
Run all-MiniLM-L6-v2 in your browser
all-MiniLM-L6-v2 downloads 28.6 MB over WebGPU (q4f16), or 21.8 MB over WebAssembly (uint8), to turn text into vectors for search and comparison entirely in the tab with Transformers.js. No install, no server.
Embeddings · Xenova/all-MiniLM-L6-v2
- Parameters
- 22.7M
- Pipeline task
- feature-extraction
Reading
The WASM build downloads smaller here: 21.8 MB (uint8) against 28.6 MB (q4f16) for WebGPU. The CPU-fallback quant compresses tighter than the GPU pick for this model.
The smallest embeddings download in the catalog (28.6 MB).
The quant ladder spans 4.0x: 21.8 MB (uint8) to 86.2 MB (fp32).
Run it in your browser
Checking what this browser can run…
All measured variants
| Variant | Size |
|---|---|
| uint8 WASM pick | 21.8 MB |
| q4f16 WebGPU pick | 28.6 MB |
| fp16 | 43.2 MB |
| q8 | 43.8 MB |
| bnb4 | 51.4 MB |
| q4 | 52.1 MB |
| fp32 | 86.2 MB |
Sizes measured from the HuggingFace API file tree for Xenova/all-MiniLM-L6-v2, not estimated.
Use it with Transformers.js
import { pipeline } from "@huggingface/transformers";
const pipe = await pipeline("feature-extraction", "Xenova/all-MiniLM-L6-v2", {
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 all-MiniLM-L6-v2, 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 all-MiniLM-L6-v2?
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.