Sparse Attention Explained: How LLMs Handle Million-Token Contexts Without Melting Your GPU
How sparse attention cuts LLM inference cost by 10x on long contexts. Covers DeepSeek NSA, MInference, H2O, and The Sparse Frontier's findings.
Deepseek
DeepSeek V4 Pro Review: 80% SWE-bench at 1/7th Claude's Price
DeepSeek V4 Pro scores 80.6% on SWE-bench Verified at $1.74/M input tokens — 7x cheaper than Claude Opus 4.7. Real benchmarks, costs, and safety gaps.
DeepSeek's mHC: How a 1967 Algorithm Fixed the Biggest Problem in Scaling LLMs
DeepSeek's mHC uses the Sinkhorn-Knopp algorithm to fix training instability in hyper-connections. Here's how doubly stochastic matrices stabilize LLM scaling.