Visualize LLM sharding performace metrics for various compute backends and LLMs in multi-node configurations using roofline model