Qwen2.5-Math-7B-Instruct-MNN

Qwen2.5-Math-7B-Instruct-MNN

Introduction

This model is a 4-bit quantized version of the MNN model exported from Qwen2.5-Math-7B-Instruct using llmexport.

Download

# install huggingface
pip install huggingface

# shell download
huggingface download --model 'taobao-mnn/Qwen2.5-Math-7B-Instruct-MNN' --local_dir 'path/to/dir'

# SDK download
from huggingface_hub import snapshot_download
model_dir = snapshot_download('taobao-mnn/Qwen2.5-Math-7B-Instruct-MNN')

# git clone
git clone https://www.modelscope.cn/taobao-mnn/Qwen2.5-Math-7B-Instruct-MNN

Usage

# clone MNN source
git clone https://github.com/alibaba/MNN.git

# compile
cd MNN
mkdir build && cd build
cmake .. -DMNN_LOW_MEMORY=true -DMNN_CPU_WEIGHT_DEQUANT_GEMM=true -DMNN_BUILD_LLM=true -DMNN_SUPPORT_TRANSFORMER_FUSE=true
make -j

# run
./llm_demo /path/to/Qwen2.5-Math-7B-Instruct-MNN/config.json prompt.txt

Document

MNN-LLM