The CRCD team has developed our homegrown generative AI chatbot model for our users. We call it ChatCRCD and we welcome users to try out the new system and let it know how it performs for you. Like all generative AI, ChatCRCD is a work in progress.
ChatCRCD is a retrieval-augmented generation (RAG) agent designed to search the CRCD user manual and provide real-time technical support, helping users navigate CRCD computing resources more easily and effectively. It is built on Meta’s open-source Llama 3.2 model, with a deployed size of 70B parameters.
The chat interface is based on Hugging Face’s open-source ChatUI, offering a user-friendly, ChatGPT-style experience with support for chat history. If you are looking for quick answers from the CRCD user manual, we encourage you to make use of this tool.
We are actively working on a new version of ChatCRCD that will incorporate additional context sources and improved mechanisms to minimize hallucinations and enhance the chat responses.
Please note that while ChatCRCD retrieves context from the CRCD user manual to answer user queries, the underlying foundational model was not specifically trained or fine-tuned on general questions about CRCD such as our mission or location data. As a result, it may still hallucinate or generate fabricated information.We recommend verifying any critical information if you use ChatCRCD for general conversation purposes.
One critical tip: the more precise the information, the more accurate the answer. A query about "University of Pittsburgh Center for Research Computing and Data" will return a more accurate response than just "Center for Research Computing."
Try ChatCRCD and see what you think: https://chatcrcd.crc.pitt.edu/. You must be within the Pitt firewall or Pitt's GlobalProtect VPN to access ChatCRCD. As always, we want feedback from our users. You can send comments to bgc14@pitt.edu.