The article also explains how to offer mathstral as a service using the GaiaNet project. After installing the required software, the model can be downloaded and run, serving over an OpenAI compatible API endpoint and a web-based chatbot UI. GaiaNet also allows users to manipulate prompts, add contexts, and incorporate a proprietary knowledge base to ground the LLM answers in truth.
Key takeaways:
- Mistral AI has released a new model called mathstral, designed specifically for math reasoning and scientific discovery. This model is finetuned and has a 32k context window.
- The model can be run locally using the LlamaEdge tech stack, and the article provides a detailed guide on how to install and use it.
- Mathstral demonstrated its ability to solve a common sense math question, proving that finetuned open source models can outperform larger closed source models in specialized areas.
- The GaiaNet project allows users to share the mathstral model as a service, adding their own context and knowledge base to ground the model's answers.