Downloads · 30 days
23
5% of all-time downloads
pj-s/Qwen3.5-REAP-212B-A17B-mxfp8
Qwen3.5-REAP-212B-A17B-mxfp8 is a machine learning model from pj-s. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as apache-2.0.
An Apple Silicon MLX Vision–Language MoE model created by applying 8-bit quantized mlx conversion to OpenMOSE/Qwen3.5-REAP-212B-A17B Router-weighted Expert Activation Pruning (REAP) model based on Qwen3.5-397B-A17B.
Downloads · 30 days
23
5% of all-time downloads
All-time downloads
435
Public
Parameters
212B
219 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors219 GB · 100%
How the weights are stored.
U32211B · 100%
From the Hugging Face model README
An Apple Silicon MLX Vision–Language MoE model created by applying 8-bit quantized mlx conversion to OpenMOSE/Qwen3.5-REAP-212B-A17B Router-weighted Expert Activation Pruning (REAP) model based on Qwen3.5-397B-A17B.
LLM runs on Mac Studio M3 Ultra+ with 256GB or 512GB unified memory.
mlx_lm.server --model pj-s/Qwen3.5-REAP-212B-A17B-mxfp8
Run this in terminal first to unlock memory restrictions:
sudo sysctl iogpu.wired_mem_breakpoint_cap=245760
(Or 250000 to be extreme), close any unused apps before launch.
mlx_vlm.convert \
--hf-path /localpath/models/OpenMOSE/Qwen3.5-REAP-212B-A17B \
--mlx-path /localpath/models/pj-s/Qwen3.5-REAP-212B-A17B-mxfp8 \
--quantize \
--q-bits 8 \
--q-group-size 32 \
--q-mode mxfp8