Wasm example - quantized embedding gemma

[vandana-rajan]

Adds a new WASM example for running quantized Embedding Gemma 300M models in the browser with WebAssembly.

Available Models:

1. Q8_0 (approx 340MB) and
2. Q4_0 (approx 297MB)

Both models are from Unsloth AI. Further, output from these models are post-processed by 2 dense layers and normalized. These are provided by Google

Demo interface follows Bert wasm example.

Changes

1. New example in candle-wasm-examples/quant-embed-gemma/
2. Modifications in candle-transformers/src/models/quantized_gemma3.rs to accommodate embedding gemma

Usage

wasm-pack build --target web --release
python3 ./serve.py --port 8000
# Opens http://localhost:8000

Note - Adding changes on top of Dr. JesseGlass's modifications to quantized_gemma3.rs

[vandana-rajan]

Hello @DrJesseGlass

I have added this new wasm example on top of your changes that is yet to be merged. Once your PR is merged, I think I can just rebase on it. Meanwhile, if you could kindly review this PR, it would be great.

Thanks,
Vandana