Add Qqp

[mianqiu00]

• Qqp

• zero-shot

  python inference.py -m Llama-2-7b-hf -d qqp -shots 0 --ranking_type prob

  Our results: 63.17. (all no)

  python inference.py -m Meta-Llama-3-8B -d qqp -shots 0 --ranking_type prob

  Our results: 62.53.

• one-shot

  python inference.py -m Llama-2-7b-hf -d qqp -shots 1 --ranking_type prob

  Our results: 36.83. (all yes)

  python inference.py -m Meta-Llama-3-8B -d qqp -shots 1 --ranking_type prob

  Our results: 69.96.

• 32-shot

  python inference.py -m Llama-2-7b-hf -d qqp -shots 32 --ranking_type prob

  Our results: 54.83.

  python inference.py -m Meta-Llama-3-8B -d qqp -shots 32 --ranking_type prob

  Our results: 68.50.

llama2 with zero-shot or one-shot predict all no or yes, this problem haven't been solved by adjusting instruction and ranking_type