Neural Memory Indexer

Researchers from Tencent, Tsinghua University, and HKUST developed FlashMemory-DeepSeek-V4, which uses Lookahead Sparse Attention to reduce memory consumption in large language models by predicting an…