## Announcements ### MatX One and our Series B [MatX One and our Series B](/content/research/series_b/index.html) - 24 Feb 2026 ### Series A [Series A](https://x.com/reinerpope/status/1899486276233146530) - 11 Mar 2025 ## Research ### Future leakage in block-quantized attention [Future leakage in block-quantized attention](/content/research/leaky_quantization/index.html) - 9 Jan 2026 ### Simple and fast Rust deriving using macro_rules [Simple and fast Rust deriving using macro_rules](/content/research/rules_derive/index.html) - 28 Jul 2025 ### Speculative Decoding with Blockwise Sparse Attention [Speculative Decoding with Blockwise Sparse Attention](/content/research/sd_nsa/index.html) - 22 Jul 2025 ### SPIRe: Boosting LLM Inference Throughput with Speculative Decoding [SPIRe: Boosting LLM Inference Throughput with Speculative Decoding](/content/research/sd/index.html) - 8 Apr 2025 ### Prioritize values over keys: faster attention with many sparsely accessed value heads [Prioritize values over keys: faster attention with many sparsely accessed value heads](/content/research/smva/index.html) - 8 Apr 2025 ### Optimize for inference too, not just training FLOPs [Optimize for inference too, not just training FLOPs](/content/research/lifetime_llm_cost/index.html) - 8 Jan 2025 ### Introducing seqax: A Simple and Efficient LLM Research Codebase [Introducing seqax: A Simple and Efficient LLM Research Codebase](/content/research/seqax/index.html) - 6 May 2024