Quantized 2_6bit MLX version of https://huggingface.co/RekaAI/reka-flash-3 that performs better than regular 2bit MLX quants with a slightly larger size. Converted using mlx-lm 0.22.2

Downloads last month: 13

Safetensors

Model size

21B params

Tensor type

U32

F16

MLX

Hardware compatibility

4-bit

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for Lowkey-Loki/reka-flash-3-mlx-2_6bit

Base model

RekaAI/reka-flash-3

Quantized

(29)

this model