RT @hingeloss: o1 style chain of thought with a local Llama 1B model (aka shrek sampler) is mostly working... hard part is intelligently p…