Hacker News new | ask | show | jobs
by viraptor 1174 days ago
This can be then quantized to the llama.cpp/gpt4all format, right? Specifically, this only tweaks the existing weights slightly, without changing the structure?