Posts

TLDR; The parallelism you chose (the way you load your matrices on GPUs) can make the difference between utilization of 50% and 90%.

TLDR; use containerization solutions instead of installing axolotl: docker or apptainer (for HPC)

This is my first blog post.