CrowdGPT - Let's train the next ChatGPT together :D

Hello, I am creating CrowdGPT, an open-source project that allows the training of a Large Language Model (LLM) in a decentralized manner. Each user can contribute their own data to enhance the AI. The concept is straightforward: rather than a single machine handling the entire training process, multiple individuals can contribute small training jobs.

These updates are then periodically merged into a shared model. The system is built around a lightweight centralized server that receives each client's training data. After receiving the data, the server merges it back into the main model. To prevent threats or malicious updates, the system employs cross-client verifications, which provide a proof of work.

Users who contribute to training the model are displayed on a leaderboard, recognizing their efforts. The data used for training is sourced from a curated dataset on Hugging Face, ensuring that no personal data is utilized during the training process. However, users are allowed to add new text to this dataset. This text undergoes moderation and validation before being incorporated.

For those interested in learning more, the GitHub repository can be found at https://github.com/Vxtzq/CrowdGPT, and the website is available at https://www.crowdgpt.net. If you have any feedback or suggestions on how to improve the project, please let me know. I am primarily seeking criticism regarding potential obstacles that might prevent users from running this on their own GPUs.

The best way to support me is by providing feedback on necessary changes to make the project fully functional, contributing to the network, starring the GitHub repository, which greatly helps the project's growth.