Please add Dflash fp16 type

[felk-dev]

There is only bf16, 8bit, 4bit options in dflash model.
and thanks for your contribution optimize omlx.
0.3.8 rc1 is almost perfect for mlx model! specprefill is amazing

[deepsweet]

Hi.

There is a PR already #880

It is working with FP16 using the first "" option already, even though it is currently saying "bf16 (default)". It's only a matter of correcting the text. See the PR for more details and benchmarks.

• https://huggingface.co/deepsweet/Qwen3.6-35B-A3B-DFlash-FP16
• https://huggingface.co/deepsweet/Qwen3.6-27B-DFlash-FP16

[felk-dev]

Wow thanks!
I already use your uploaded model😆