I’ve been using Ollama with the Mistral model in our Retrieval-Augmented Generation (RAG) setup for the past three months. While it worked well overall, it struggled with arithmetic calculations sometime only.
Recently, we implemented Mistral chaining with Mathstral, and after installation, we initially achieved accurate and reliable results across all prompts in our RAG. However, after 24 hours of testing, the model began hallucinating, failing to extract accurate information, and producing incorrect arithmetic calculations.
Could you clarify if this is a preexisting issue with Mathstral or something new? Additionally, please provide guidance on how we can resolve this problem.
Thank you!
1 Like