semioz.com

I’m Semih!

Lately I work on reinforcement learning/post-training and inference systems. Right now I’m an RL research resident at Prime Intellect, building robotics environments.

2 posts · updated may 2026

RECENT WRITINGS

Putting a model in charge of KV-cache memory Building an RL environment to learn how KV-cache eviction works in LLM serving systems rl · may 04, 2026 Speeding up diffusion models with first block caching How to speed up diffusion inference with minimal quality loss using first block caching diffusion · aug 13, 2025