Why do 23 billion active parameters not mean Beam fits in 128 gigabytes?
Active parameters describe the subset used for a token in a sparse mixture-of-experts model. The remaining experts still need to be stored or fetched somehow, and memory is also used by runtime components and the context cache. Exact requirements depend on format, software and placement strategy.
Answered in
Reflection Beam Is Open-Weight. Your Deployment Still Has a Memory Problem.Beam’s Apache-licensed weights widen access, but 501 billion total parameters make hosting, memory and evaluation the real decisions.
Read the full analysisOther questions this article answers
More how ai actually works questions
- What is Meta’s Your Algorithm feature?
- Does natural-language feedback give users full control of Facebook ranking?
- Did Facebook have no feed controls before this feature?
- What is NVIDIA RTX Spark?
- Does RTX Spark make AI development private by default?
- Will all x86 developer tools run on the ARM-based system?
- What does open-weight mean for Reflection Beam?
- Can a developer run Beam locally?
Every answer on Crashtech is written by the editor of the article it comes from — never auto-summarised. Browse all answers or the How AI Actually Works beat.