Kimi K2.7-Code: open-source coding model with better token efficiency

Transformers Are Inherently Succinct

Do transformers need three projections? Systematic study of QKV variants

Soul Player C64 – A real transformer running on a 1 MHz Commodore 64

Attention Residuals

Building a Minimal Transformer for 10-digit Addition

FORTH? Really!?

Attention at Constant Cost per Token via Symmetry-Aware Taylor Approximation

David Patterson: Challenges and Research Directions for LLM Inference Hardware