A Practical Sparse Attention Method for Long-Context LLM Inference - View it on GitHub