#inference

• [0007]2026-07-02Writing A Megakernel For LLM Decode - A Worklog#cuda#megakernel#inference#blackwell#sm120#rtx#5080
• [0006]2026-05-078.5x Faster Speech-to-Text: From 429ms to 50ms on a Single GPU#inference#stt#fp8#hopper#sm90

all posts →