Announcements

• MatX One and our Series B - 24 Feb 2026
• Series A - 11 Mar 2025

Research

• Future leakage in block-quantized attention - 9 Jan 2026
• Simple and fast Rust deriving using macro_rules - 28 Jul 2025
• Speculative Decoding with Blockwise Sparse Attention - 22 Jul 2025
• SPIRe: Boosting LLM Inference Throughput with Speculative Decoding - 8 Apr 2025
• Prioritize values over keys: faster attention with many sparsely accessed value heads - 8 Apr 2025
• Optimize for inference too, not just training FLOPs - 8 Jan 2025
• Introducing seqax: A Simple and Efficient LLM Research Codebase - 6 May 2024