BrownieSquad
We started with a surrogate discovery env and foudn that it is excellent and beats pytorch.compile (36x) and trioton.autotune (3x) using a gaussian process optimized by out oracle.
Core trick
Uses a Gaussian process oracle to pick kernel configs, then trains an Unsloth LoRA model with DPO on the winners.
Tags
Description
We started with a surrogate discovery env and foudn that it is excellent and beats pytorch.compile (36x) and trioton.autotune (3x) using a gaussian process optimized by out oracle.
Since the oracle was so powerful we used it for generative CUDA kernel optimization wth a online self-improvement loop using unsloth lora and DPO.
The linked repo returns 404, so it was deleted or made private.