Tutorials

LLM training is built on fast MatMuls. But many surrounding ops still run as memory-bound kernels. CODA reparameterizes them to hide in the …

LLM training is built on fast MatMuls. But many surrounding ops still run as memory-bound kernels. CODA reparameterizes them to hide in the matmul’s shadow, fused into its epilogue before results leav

DGX agentx-post
tutorialsjeremy-howard--x

LLM training is built on fast MatMuls. But many surrounding ops still run as memory-bound kernels. CODA reparameterizes them to hide in the matmul’s shadow, fused into its epilogue before results leave the chip. Bonus: LLMs can write fast CODA kernels too (approaching SoLs).

Source: Jeremy Howard (X) | 2026-05-21

Loading related sources…