Coming soon
If you are unable to access the GitHub website, you can open the hosts file with Notepad at C:\Windows\System32\drivers\etc\hosts (for Android system) or at etc/hosts (for Linux/Mac). At the end of the file, add the following lines:
20.205.243.166 github.com
20.205.243.166 www.github.com
20.205.243.166 api.github.com
20.205.243.166 assets-cdn.github.com
20.205.243.166 raw.githubusercontent.com
20.205.243.166 gist.github.com After saving and closing the file, refresh the DNS cache by opening the terminal and running:
For Windows:
ipconfig /flushdns For Linux/Mac:
sudo systemd-resolve --flush-caches Create the conda environment. We recommend using different conda environments to avoid conflicts, such as "internlm", "llama" and "qwen". Here is an example.
conda_env_name=llama
conda create -n $conda_env_name python=3.10.0 -y
conda activate $conda_env_nameAfter setting up the environment, some packages are necessary to be installed. We provide a requirements.txt. Please be careful about the version of these packages.
cd PIC_Project
pip install -r requirements.txtThen, download the model checkpoints using the command . Make sure that you have suffient hard drive space. (Ensure that the path to the downloaded model contains only English characters.)
python download.pyWe provide two ways to conduct LLMs' reasoning, i.e., online inference and API call.
Change the graphics card number and python file to utilize different models.
cd inference
CUDA_VISIBLE_DEVICES=0
python llama_online.pyYou should suspend the API service. This process may take ~30 seconds. Check whether the API service is successfully started in the corresponding log file in the "PIC_Project/log" folder.(Ensure that the server IP address of the computer has been modified.)
cd api_setup
bash llama_api.shSubsequently, use the conmand below to start your chat.
cd inference
python llama_offline.pyPIC_Project is carried out by six students of ECPK in Beihang University. For questions, please contact Linqing Zhong or Zhuyun Yuan.