how to use --instruction_tokens?

[loubnabnl]

This argument is only used for InstructHumanEval benchmark and its use is explained in this documentation. For example:

accelerate launch  main.py \
  --model <MODEL_NAME> \
  --max_length_generation <MAX_LENGTH> \
  --tasks instruction-humaneval \
  --instruction_tokens <user_token>,<end_token>,<assistant_token>\
  --temperature 0.2 \
  --n_samples 50 \
  --batch_size 50 \
  --allow_code_execution