I heard that ChatGLM3 was released. Zhang Xiaobai wanted to try it on Windows:
Install Git Large File Storage first:
Open git-lfs.github.com
Click Download
Install the downloaded file:
Open Git Bash:
Execution: git lfs install
At this point, Git LFS is installed successfully.
cd /f/
git clone https://huggingface.co/THUDM/chatglm3-6b
For well-known reasons, this road is blocked.
I can only take a look at it from modelscope:
Open https://modelscope.cn/models/ZhipuAI/chatglm3-6b/summary
Open the Anaconda PowerShell prompt
conda activate model310
F:
Enter the python command line and execute:
from modelscope import snapshot_download
model_dir = snapshot_download("ZhipuAI/chatglm3-6b", revision = "v1.0.0")
The downloaded model file is placed in C:\Users\xishu\.cache\modelscope\hub\ZhipuAI\chatglm3-6b
Move it to F:\models\THUDM\chatglm3-6b
Back to F:\
git clone https://github.com/THUDM/ChatGLM3
I have no choice but to download the zip package and unzip it:
cd F:\ChatGLM3-main
conda deactivate
conda create -n chatglm3 python=3.10
conda activate chatglm3
pip install -r requirements.txt -i https://pypi.tuna.tsinghua.edu.cn/simple
The CPU version of torch 2.1.0 must be installed here. Remember to use conda to change to the GPU version.
Open
according to:
conda install pytorch torchvision torchaudio pytorch-cuda=11.8 -c pytorch -c nvidia
pip install chardet
Execute the following script on the Python command line:
from transformers import AutoTokenizer, AutoModel
tokenizer = AutoTokenizer.from_pretrained("\models\THUDM\chatglm3-6b", trust_remote_code=True)
model = AutoModel.from_pretrained("\models\THUDM\chatglm3-6b", trust_remote_code=True, device='cuda')
model = model.eval()
response, history = model.chat(tokenizer, "Hello", history=[])
print(response)
response, history = model.chat(tokenizer, "What can you do", history=history)
print(response)
response, history = model.chat(tokenizer, "Tell me what kind of movie Forrest Gump is", history=history)
print(response)
pip install jupyter -i https://pypi.tuna.tsinghua.edu.cn/simple
pip install jupyterlab -i https://pypi.tuna.tsinghua.edu.cn/simple
cd composite_demo
Modify client.py
Will
MODEL_PATH = os.environ.get(‘MODEL_PATH’, ‘THUDM/chatglm3-6b’)
Change to
MODEL_PATH = “\models\THUDM\chatglm3-6b”
streamlit run main.py
The browser automatically pops up:
Switch to Tool:
Let’s try to see if ChatGLM3 can be embedded into LongChain-ChatChat.
Open the configuration file F:\Langchain-Chachat\configs\model_config.py
Add the following content:
Switch to chatchat conda environment
conda deactivate
conda activate chatchat
F:
cd \Langchain-Chachat
python startup.py –all-webui
After the window pops up, switch to LLM conversation mode:
It can actually display Latex formulas.
Baidu looked at the equation:
Is this the right answer?
(Full text finished, thank you for reading)