LLM Implementation Weekly
A weekly filter for edge AI and LLM inference & deployment · Papers / Releases / Tools / Opinions · signal only, no noise
Issue 8: llama.cpp 1.0 and the Vera Rubin Week
Nov 11 – Nov 17, 2026 · 5 picks · 5 quick takes
Issue 7: Inference Control Planes and the FP4 Kernel War
Nov 4 – Nov 10, 2026 · 7 picks · 6 quick takes
Issue 6: The Reproducibility War in Quantization
Oct 28 – Nov 3, 2026 · 6 picks · 5 quick takes
Issue 5: Multi-Tenant KV Safety and Non-Portable Speedups
Oct 21 – Oct 27, 2026 · 7 picks · 5 quick takes
Issue 4: The Heterogeneous-Hardware Sprint and Budget-Aware KV Compression
Oct 14 – Oct 20, 2026 · 7 picks · 5 quick takes
Issue 3: Rubin Day-0 and the KV-Compression Engineering Turn
Oct 7 – Oct 13, 2026 · 7 picks · 5 quick takes
Issue 2: The Fall Release Wave of Inference Engines
Sep 30 – Oct 6, 2026 · 7 picks · 5 quick takes
Issue 1: The Cost-of-Intelligence War
Sep 22 – Sep 29, 2026 · 10 picks · 14 quick takes
About this column
This column tracks the full GPU-to-Token implementation chain: inference framework releases (vLLM / SGLang / llama.cpp), new papers on inference optimization and model compression, edge hardware progress, and hands-on community benchmarks. The column launched in week 40 of 2026, so Issue 1 is the founding issue.
Editorial policy: entries with measured data or code come first, vendor marketing is down-weighted, and anything we cannot verify is tagged [Unverified] and kept out of the picks.