VoxCPM

VoxCPM

This version of VoxCPM has been converted to run on the Axera NPU using w8a16 quantization. Compatible with Pulsar2 version: 4.2

Convert tools links:

For those who are interested in model conversion, you can try to export axmodel through the original repo : VoxCPM offical

Pulsar2 Link, How to Convert LLM from Huggingface to axmodel

AXera NPU HOST LLM Runtime

Support Platform

• AX650
  • AX650N DEMO Board
  • M4N-Dock(爱芯派Pro)
  • M.2 Accelerator card

How to use

Download all files from this repository to the device

1. Run python demo

1. Install voxcpm axinfer package

git clone -b 1.0.4-axmode_infer https://github.com/techshoww/VoxCPM.git 
cd VoxCPM
pip3 install .

2. Download zipenhancer

pip3 install modelscope  
modelscope download --model iic/speech_zipenhancer_ans_multiloss_16k_base --local_dir iic/speech_zipenhancer_ans_multiloss_16k_base

3. Run on Axera Device

Go to the root directory of this project. run:

python3 run_ax650.py

2. Run c++ demo

1. Install transformers

pip3 install transformers>=4.56.2

2. Start tokenizer server

python3 tokenizer.py --port 9999

3. Run c++ demo

bash run_ax650.sh