Downloads · 30 days
27
6% of all-time downloads
matvei-aleksandrovich/Qwen3.8-27B-MTP-4bit
Qwen3.8-27B-MTP-4bit is a text generation model from matvei-aleksandrovich. Use it when you need the model to write or continue text. It is set up for mlx. The card lists the license as apache-2.0.
This repository contains the native Multi-Token Prediction weights extracted from Qwen/Qwen3.8-27B and quantized to 4-bit affine MLX weights.
Downloads · 30 days
27
6% of all-time downloads
All-time downloads
438
Public
Parameters
425M
252 MB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors239 MB · 92%
How the weights are stored.
U32425M · 100%
From the Hugging Face model README
This repository contains the native Multi-Token Prediction weights extracted
from Qwen/Qwen3.8-27B and
quantized to 4-bit affine MLX weights.
It is not a standalone language model. Use it only as the MTP drafter for
matvei-aleksandrovich/Qwen3.8-27B-MLX-4bit.
qwen3_5_mtpmtperokhins/mlx-vlm, based on commit 77f16c25python -m mlx_vlm.generate \
--model matvei-aleksandrovich/Qwen3.8-27B-MLX-4bit \
--draft-model matvei-aleksandrovich/Qwen3.8-27B-MTP-4bit \
--draft-kind mtp \
--prompt "Write a short Python example." \
--max-tokens 256
The target model verifies every proposed token. Measure the acceptance rate and end-to-end speed on representative prompts before production use.