Added harfmulness eval results.

[TheRootOf3]

Improved the evaluation runner script eval_harmfulness/run_experiment.sh.

Added evaluation results for the following models:

• All checkpoints (0 - 1000 every 100) of the ByteDance reproduction models using 5 seeds:
  • opt1.3b_unlearned_harmful-for-real42-eval1
  • opt1.3b_unlearned_harmful-for-real1234-eval1
  • opt1.3b_unlearned_harmful-for-real456-eval1
  • opt1.3b_unlearned_harmful-for-real42-eval1
  • opt1.3b_unlearned_harmful-for-real114514-eval1
• test-batch-eval1 - A test evaluation of a batch unlearning with sample size = 32, batch size = 8 and num epochs = 20.
• test-seq-eval2 - A test evaluation of sequential unlearning with sample size = 32, sequential = 4, batch size = 8 and num epochs = 20.