random-prompt

Code and supplementary document for paper Prompt Optimisation with Random Sampling

Most up-to-date version Strings from the Library of Babel: Random Sampling as a Strong Baseline for Prompt Optimisation

References

If you use this repository in your research, please cite our paper:

@inproceedings{lu-etal-2024-strings,
    title = "Strings from the Library of Babel: Random Sampling as a Strong Baseline for Prompt Optimisation",
    author = "Lu, Yao  and
      Wang, Jiayi  and
      Tang, Raphael  and
      Riedel, Sebastian  and
      Stenetorp, Pontus",
    editor = "Duh, Kevin  and
      Gomez, Helena  and
      Bethard, Steven",
    booktitle = "Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)",
    month = jun,
    year = "2024",
    address = "Mexico City, Mexico",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.naacl-long.122",
    pages = "2221--2231",
}

Implementation from scratch

Want to implement from scratch? You can take a look at the core implementation for generating random separators in less than 10 lines of code.

1. Random vocabulary mode

import random
from transformers import GPT2Tokenizer

prompt = "this is a good movie [Answer:] positive"

tokenizer = GPT2Tokenizer.from_pretrained("gpt2")
vocab_size = tokenizer.vocab_size

# random length for separator
separator_length = random.randint(1, 5)
random_separator_ids = random.sample(range(vocab_size), separator_length)
random_separator_text = tokenizer.decode(random_separator_ids, skip_special_tokens=True)

random_prompt = prompt.replace("[Answer:]", random_separator_text)

# evaluate on training set
# ...

2. Random without context mode

from transformers import GPT2Tokenizer, GPT2LMHeadModel

prompt = "this is a good movie [Answer:] positive"

model = GPT2LMHeadModel.from_pretrained('gpt2')
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")

# random length for separator
separator_length = random.randint(1, 5)
random_separator_ids = model.generate(do_sample=True, max_new_tokens=separator_length)[0]
random_separator_text = tokenizer.decode(random_separator_ids)

random_prompt = prompt.replace("[Answer:]", random_separator_text)

# evaluate on training set
# ...

3. Random with context mode

from transformers import GPT2Tokenizer, GPT2LMHeadModel

prompt = "this is a good movie [Answer:] positive"

# follow OPRO's examples format https://arxiv.org/abs/2309.03409
context = "I like this movie <INS> positive\nI don't like this movie <INS>\n"

model = GPT2LMHeadModel.from_pretrained('gpt2')
tokenizer = GPT2Tokenizer.from_pretrained("gpt2")

# random length for separator
separator_length = random.randint(1, 5)
context_input_ids = tokenizer.encode(context, return_tensors='pt')
random_separator_ids = model.generate(context_input_ids, do_sample=True, max_new_tokens=separator_length)[0]
random_separator_text = tokenizer.decode(random_separator_ids)

random_prompt = prompt.replace("[Answer:]", random_separator_text)

# evaluate on training set
# ...