Strata boosts decode speed from 70–80 tk/s to 150 tk/s for qwen 3.8 flash next on llamacpp; user reports DeepSeek v4.1flash 2.25bit achieves 70tk/s decode

Have been using qwen 3.8 flash next with llamaserver. Was getting 70-80tk/s decode and was happy. Then read about strata and decided to give it a try. Decode shot up to 150 tk/s. Not sure what magic it's using? You...