Comment by retinaros

its not that loose of a version. its the reality and as probably is surely a focus of a dedicated post training RL-ing these kind of githubs. of course you would train specifically on the task. you would mix this eval data with others in thousands of githubs repos.