Microsoft makes new 1.3B coding LLM that outperforms all models on MBPP except GPT-4, reaches third place on HumanEval above GPT-3.5, and shows emergent properties
