Search for a command to run...
On the importance of pre-training data volume for compact language models