SSSAMMMM
/

Qwen3.8-27B-DFlash2-MLX-4bit

speculative-decoding

4-bit precision

Model card Files Files and versions

Qwen3.8-27B-DFlash2 (MLX 4-bit)

4-bit MLX export of z-lab/Qwen3.8-27B-DFlash2, the DFlash2 speculative-decoding draft head for Qwen/Qwen3.8-27B.

Quantization: affine, 4 bits, group size 64 (config.json → quantization).
Use as the draft model next to a Qwen3.8-27B MLX target with mlx-vlm (--draft-kind dflash).
Weights are unchanged apart from quantization. License follows the upstream Apache-2.0 release.

Downloads last month: 206

Safetensors

Model size

2B params

Tensor type

U32

·

BF16

·

MLX

Hardware compatibility

Log In to add your hardware

4-bit

Inference Providers NEW

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

Model tree for SSSAMMMM/Qwen3.8-27B-DFlash2-MLX-4bit

Base model

Qwen/Qwen3.8-27B

Finetuned

z-lab/Qwen3.8-27B-DFlash2

Quantized

(19)

this model