Minimal clean reproduction of n-gram value memory induced replay-specific train/val gap on vanilla nanoGPT
These files are guoshaoyang-pku/ngram-gap-lab's own configuration. They tell Codex how to work on this repository, so they are not mods to install elsewhere. Copy one as a starting point and replace the parts that are about this project.
.agents/skills/ngram-gap-experiment-registration/SKILL.md A 66 tok .agents/skills/ngram-gap-plotting/SKILL.md A 67 tok .agents/skills/ngram-gap-settings/SKILL.md A 54 tok