Initial BF16 bundle: Gemma-4-E4B BF16 + MERaLiON speech LoRA, 16.09% WER

83de0b8 verified 16 days ago

432 Bytes

	# Provenance

	This repository is an unquantized MLX conversion of `google/gemma-4-E4B-it`.

	\| Item \| Value \|
	\|---\|---\|
	\| Upstream model \| `google/gemma-4-E4B-it` \|
	\| Conversion target \| MLX safetensors \|
	\| Dtype \| bfloat16 \|
	\| Quantization \| none \|
	\| Converter \| `mlx_lm.convert` \|

	No fine-tuning, merging, pruning, or weight quantization was applied. The conversion preserves the upstream BF16 numerical format for MLX inference.

	# Provenance

	This repository is an unquantized MLX conversion of `google/gemma-4-E4B-it`.

	\| Item \| Value \|
	\|---\|---\|
	\| Upstream model \| `google/gemma-4-E4B-it` \|
	\| Conversion target \| MLX safetensors \|
	\| Dtype \| bfloat16 \|
	\| Quantization \| none \|
	\| Converter \| `mlx_lm.convert` \|

	No fine-tuning, merging, pruning, or weight quantization was applied. The conversion preserves the upstream BF16 numerical format for MLX inference.