Comment by peakji

It is an LLM fine-tuned using a new type of dataset and RL reward. It's good at reasoning, but I would not recommend to replace Llama for general tasks.