Skip to content
localmodel.run

Runs in the browser · Reranking

Run BGE Reranker v2 M3 in your browser

BGE Reranker v2 M3 downloads 669.6 MB over WebGPU (q4f16), or 544.3 MB over WebAssembly (uint8), to score how well a passage answers a query entirely in the tab with Transformers.js. No install, no server.

Reranking · onnx-community/bge-reranker-v2-m3-ONNX

WebGPU download
669.6 MB
q4f16
WASM download
544.3 MB
uint8
Parameters
567.76M
Pipeline task
text-classification

Reading

Rerankers score a (query, passage) pair, not a single string. Confirm the model card's expected input shape before assuming the generic text-classification pipeline() call handles pairs; AutoModelForSequenceClassification with a tokenizer pair encoding is the safer default.

The WASM build downloads smaller here: 544.3 MB (uint8) against 669.6 MB (q4f16) for WebGPU. The CPU-fallback quant compresses tighter than the GPU pick for this model.

The quant ladder spans 4.0x: 544.3 MB (uint8) to 2.12 GB (fp32).

Will it run in your browser?

Live check this browser

Checking for WebGPU support…

All measured variants

Quant Download size
Variant Size
uint8 WASM pick 544.3 MB
q4f16 WebGPU pick 669.6 MB
fp16 1.06 GB
q8 1.06 GB
bnb4 1.15 GB
q4 1.17 GB
fp32 2.12 GB

Sizes measured from the HuggingFace API file tree for onnx-community/bge-reranker-v2-m3-ONNX, not estimated.

Use it with Transformers.js

import { pipeline } from "@huggingface/transformers";

const pipe = await pipeline("text-classification", "onnx-community/bge-reranker-v2-m3-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

FAQ

WebGPU or WASM for BGE Reranker v2 M3, 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 BGE Reranker v2 M3?

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.