mlx-community/Nemotron-3-Embed-1B-BF16-4bit

Record source revision, artifact sha256 and embedding L2 norm

Make the comparison script self-contained

Fix MIRACL tables missing header rows

Restore Usage and Repository contents sections

Note that MIRACL is in the upstream model's training data

Scope the card to conversion fidelity and quantization impact

Remove non-standard, non-reproducible KLUE-derived results

State the confidence interval on the Korean check

Document measured variant tradeoff: quantized is slower, not faster

Add the backend comparison script

Reference the bundled benchmark harness

Add the benchmark harness used for the card numbers

MLX conversion of nvidia/Nemotron-3-Embed-1B-BF16

initial commit