Teaching a 27B Model to Write Triton Kernels: SFT + RL Gym

In Part 1, we explored how GPU kernels work under the hood — why operator fusion and memory tiling matter, and how benchmarks like KernelBench evaluate whether an LLM can write kernels that actually b