Instruction based dataset used to instruct-tune Codellama model for APR based tasks
Three fields:
Instruction: One of the chosen 5 instructions
Context: The context/input supplied along with the instruction. In this case, it consists of
PREFIX CODE SUFFIX CODE which are the prefix, suffix and middle tokens for Codellama but can be replaced with other tokens as well depending on model.
Response: The predicted response (which is the middle missing part in the code)
