Huge bit chasing effort getting colqwen2 into llama.cpp this evening. Mostly around the image processing stack. Not going to be doing in process embedding of images much longer.
Thomas Woodfeels good. covering a lot of ground. Impatient to get started on ACP version 2 rewrite. I just put LFM2.5 ColBERT inside llama.cpp, then ollama because I want their model loading management/maintenance and enormous userbase reliability.