Oct 19, 2026 Diffusion LLMs: what changes when you stop generating one token at a time Aug 29, 2026 Serving LLMs: what actually decides your latency