Question: Feasibility of using more modern small language models

Hi I saw the reddit blog post on claudish and thought this is a very interesting thing. I was thinking if using a newer model like LFM2.5-230M or LFM2.5-350M might help to drastically improve the model's capabilities? I understand that changing models would require recompilation of the .paw file. I also saw some things about Cactus' runtime being made for edge and browser use, not sure if its is a promising alternative to what is used now. Also I would like to know more about the compilation process on the server as I dont think the source is published? It would help with experimentation greatly. Im also thinking about the possibility to use this as part of a RAG? What do you think? Thank you, just wanted to share some thoughts I had on this.

The issue names no files, tests, or entry points. Start by reviewing the repository's compilation and runtime entry points, then determine whether model selection, Cactus runtime support, server-side compilation, or RAG integration is in scope and what concrete outcome would define completion.