conda create -n torch2.0 python=3.9
conda activate torch2.0
pip install transformers==4.26.1 icetk==0.0.5 cpm-kernels==1.0.11 protobuf==3.19.5 gradio==3.22.1 streamlit==1.20.0 streamlit-chat==0.0.2.2
mkdir chatGLM_6B
cd chatGLM_6B
# Huggingface (3G)
sudo apt-get install git-lfs
git init
git lfs install
git clone https://huggingface.co/THUDM/chatglm-6b-int4-qe
# Baidunetdisk (3G)
链接: https://pan.baidu.com/s/1D_-ZHWcyCrkwXus5FmuK1g?pwd=m2si
提取码: m2si
conda activate stablediffusion
pip install torch==1.12.1+cu113 --extra-index-url https://download.pytorch.org/whl/cu113
pip install transformers==4.19.2 diffusers==0.12.1 invisible-watermark
# -------- tree -h -----------
chatglm-6b-int4-qe/
├── [ 797] config.json
├── [4.0K] configuration_chatglm.py
├── [2.6M] ice_text.model
├── [ 11K] LICENSE.txt
├── [ 51K] modeling_chatglm.py
├── [2.3K] MODEL_LICENSE.txt
├── [2.9G] pytorch_model.bin
├── [1.1K] quantization_kernels.c
├── [1.6K] quantization_kernels_parallel.c
├── [ 28K] quantization.py
├── [4.6K] README.md
├── [ 12K] tokenization_chatglm.py
└── [ 416] tokenizer_config.json
# ---------------------------
python demo.py
python cli_demo.py
python web_demo.py
python web_demo2.py
- deploy && inference test
- fine-tune
https://github.com/THUDM/ChatGLM-6B
https://huggingface.co/THUDM/chatglm-6b-int4-qe