Downloads · 30 days
21
16% of all-time downloads
goodman20241017/Qwen3.5-35B-A3B-BPD2-g128
Qwen3.5-35B-A3B-BPD2-g128 is a text generation model from goodman20241017. Use it when you need the model to write or continue text. The card lists the license as apache-2.0.
This model is a 2-bit quantized version of Qwen/Qwen3.5-35B-A3B using the BPDQ (Bit-Plane Decomposition Quantization) algorithm.
Downloads · 30 days
21
16% of all-time downloads
All-time downloads
134
Public
Parameters
4.5B
13.2 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors13.2 GB · 100%
How the weights are stored.
I322.1B · 47%
From the Hugging Face model README
This model is a 2-bit quantized version of Qwen/Qwen3.5-35B-A3B using the BPDQ (Bit-Plane Decomposition Quantization) algorithm.
BPDQ is a post-training quantization (PTQ) method that constructs a variable quantization grid via bit-planes and scalar coefficients. Unlike conventional methods that use fixed uniform intervals, BPDQ iteratively refines the grid using second-order information (Hessian-induced geometry) to minimize output discrepancy. This allows for high-fidelity performance even in ultra-low bit regimes (2-3 bits).
| Model | SIZE | Wiki | Arc_C(norm) | HellaS(norm) | Math500(verify) |
|---|---|---|---|---|---|
| Qwen3.5-35B-A3B | 67G | 7.46 | 61.77 | 82.37 | 54.6 |
| Qwen3.5-35B-A3B-bpdq-W2G128 | 13G | 10.96 | 59.47 | 75.14 | 34.2 |
Note: This is a preliminary experimental version and has not yet been specifically optimized for Mixture-of-Experts (MoE) models.
To use this model, you need to apply the BPDQ patch to GPTQModel (v5.7.0) as described in the official GitHub repository.
@article{chen2026bpdq,
title={BPDQ: Bit-Plane Decomposition Quantization on a Variable Grid for Large Language Models},
author={Chen, Junyu and Li, Jungang and Xiong, Jing and Wang, Wenjie and Yang, Qingyao and Xiao, He and Li, Zhen and Wu, Taiqiang and Chen, Mengzhao and Peng, Zhen and others},
journal={arXiv preprint arXiv:2602.04163},
year={2026}
}