Downloads · 30 days
0
Qwen3.6-27B-exl3-2.00bpw
LiRiu/Qwen3.6-27B-exl3-2.00bpw
Qwen3.6-27B-exl3-2.00bpw is a machine learning model from LiRiu. Use it for the machine learning task on the model card, and read the license before you ship it in a product.
Access
Public
Updated Jul 16, 2026
Repo size
—
Likes
0
Public
Repo makeup
Click a slice to open those files.
Other1.5 KB · 56%
At a glance
- Access
- Public
- Created
- Jul 15, 2026
- Updated
- Jul 16, 2026
- SHA
- 5fe6e6f2
Model card
From the Hugging Face model README
Qwen3.6-27B EXL3 2.00 bpw
量化实验记录。网络环境所限，模型权重未上传，仅保留实验环境配置。
实验配置
| 项目 | 值 |
|---|---|
| 基座模型 | Qwen/Qwen3.6-27B |
| 量化格式 | EXL3 (ExLlamaV3 v1.0.0) |
| 平均位宽 | 2.00 bpw |
| GPU | NVIDIA A100 |
环境
| 项目 | 值 |
|---|---|
| OS | Ubuntu 20.04.3 LTS |
| Kernel | 5.15.0-56-generic |
| GCC | 12.4.0 (conda-forge) |
| NVIDIA Driver | 570.211.01 |
| NVCC | 12.4.131 |
| Python | 3.10.20 |
| PyTorch | 2.5.1+cu121 |
| Triton | 3.1.0 |
| Flash-Attn | 未安装（triton 正常） |
已知问题
以下修改是使 ExLlamaV3 v1.0 + Triton 3.1 正常工作的必要改动：
- 布尔链语法：
triton_paged.py中if X and not Y and not Z需拆分中间变量绕过 Triton 3.1 限制 - Triton API：
bc_attn.py中constexprs=改为constants= - NVCC 路径：conda env
CUDA_HOME需正确指向$CONDA_PREFIX，CPLUS_INCLUDE_PATH设置targets/x86_64-linux/include
量化命令
EXL3_PREFER_FA2=1 python convert.py \
-i /root/qwen3.6-27b \
-o /root/qwen3.6-27b-exl3-2.00bpw \
-w /root/temp/qwen-exl3 \
-b 2.00
- Created
- Jul 15, 2026
- Updated
- Jul 16, 2026