GongBU

Bolin97/GongBU is ranked 11,077 among these 8 peers; SakanaAI/self-adaptive-llms is ranked 11,073.

Once deployed, there is no need to write any code to finetune a model, and the platform can be used by anyone with a web browser. The platform is built on top of the Transformers, Peft, etc. and is designed to be easy to use for non-technical users. 📈 You can easily finetune, evaluate, and deploy models with the platform. 📦 Currently, by default, the platform should be started with docker-compose in a linux system with nvidia-docker. However, by

installing the required dependencies, with correct configurations, the platform can be run natively. 📦 Recommended Way of Installation The recommended installation is to use the docker-compose file provided in the repository on a linux system with nvidia-docker. This is the easiest way to install the platform. It's simple, clone the repository, then run the (inquirer and rich libraries are required) to download necessary files that are not included in the repository, micromamba and a bert model to be exact

. You need to install docker and nvidia-docker. Then run the following command: The platform should be running on behind an nginx reverse proxy. 🔧 Usage After accessing the platform, you will be welcomed by the log in page. By default, the platform does not provide any user, so you need to create a user by clicking the sign up button. When signing up, by default, you will need a token to sign up. It will be in the file in

the root directory of the project. You can change the token in the however you want. Nevertheless, you can use a permissive registering strategy by setting the environment variable in the docker compose file, which will allow anyone to sign up. After logging in, you will find everything to be empty, here's how you can import assets into the platform, only two kinds of assets are needed, the dataset and the model. To upload a dataset, you need to go

to the data page. The datasets are arranged in pools, with each pools containing several datasets. You can create a new pool by clicking the create button, and follow the instructions to upload a dataset. You can also click the details button of an existing pool to upload a dataset to that pool. To download models, you need to go to the model download page. After you have uploaded the datasets and downloaded the models, you can fintune, evaluate and

deploy the models on respective pages. ✅ Current Status The platform is now fully functional, albeit under constant development and polishing, especially the frontend, and breaking changes may take place. 🧾 Trivial We used bun instead of node, which is a blazingly-fast, all-in-one js bundler, runtime, test runner, package manager, etc. It is a relatively new project, but its performance attracted us to use it. However, if you want to switch to node for better stability, simply change all

the bun-related command in the dockerfile to node or npm. ⚠️ Disclaimer FOR ANY LOSS OR DAMAGE, THE DEVELOPER IS