I have 2 use cases 1. I want to cache the system prompt across multiple invocations 2. I want to upload the file to gemini and use that across prompts to get token benefits. Are these possible through BAML?

If you use the modular API this is possible! https://docs.boundaryml.com/guide/baml-advanced/modular-api Unfortunately we only native support this for Anthropic right now, but we should definitely support this for Gemini