Hugging Face has introduced a new leaderboard for code language models, called LiveCodeBench. This evaluation framework assesses the performance of code LLMs in a holistic and contamination-free manner. The leaderboard aims to provide a more comprehensive understanding of these models' capabilities by evaluating their ability to complete tasks without relying on pre-existing code or external libraries. --- Why it matters: This matters to researchers and engineers working with code LLMs because it provides a standardized way to compare the performance of different models, which can help identify strengths and weaknesses in their design. This can inform the development of more effective and efficient code generation capabilities. Source: https://huggingface.co/blog/leaderboard-livecodebench