metadata

annotations_creators:
  - machine-generated
language:
  - en
language_creators:
  - machine-generated
  - expert-generated
license:
  - mit
multilinguality:
  - monolingual
pretty_name: MultiPLE-E
size_categories:
  - 1K<n<10K
source_datasets:
  - original
  - extended|openai_humaneval
tags: []
task_categories: []
task_ids: []

Dataset Card for MultiPL-E

Dataset Description

Homepage: https://nuprl.github.io/MultiPL-E/
Repository: https://github.com/nuprl/MultiPL-E
Paper: https://arxiv.org/abs/2208.08227
Point of Contact: [email protected], [email protected], [email protected]

Dataset Summary

MultiPL-E is a dataset for evaluating large language models for code generation that supports 18 programming languages. It takes the OpenAI "HumanEval" Python benchmarks and uses little compilers to translate them to other languages. It is easy to add support for new languages and benchmarks.

Example

The following script uses the Salesforce/codegen model to generate Lua and MultiPL-E to produce a script with unit tests for luaunit.

import datasets
from transformers import AutoTokenizer, AutoModelForCausalLM

LANG = "lua"
MODEL_NAME = "Salesforce/codegen-350M-multi"
tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME)
model = AutoModelForCausalLM.from_pretrained(MODEL_NAME).half().cuda()
problems = datasets.load_dataset("nuprl/MultiPL-E", LANG)

def stop_at_stop_token(decoded_string, problem):
    """
    Truncates the output at stop tokens, taking care to skip the prompt
    which may have stop tokens.
    """
    min_stop_index = len(decoded_string)
    for stop_token in problem["stop_tokens"]:
        stop_index = decoded_string.find(stop_token)
        if stop_index != -1 and stop_index > len(problem["prompt"]) and stop_index < min_stop_index:
            min_stop_index = stop_index
    return decoded_string[:min_stop_index]

for problem in problems["test"]:
    input_ids = tokenizer(
        problem["prompt"],
        return_tensors="pt",
    ).input_ids.cuda()
    generated_ids = model.generate(
        input_ids, max_length=256, pad_token_id=tokenizer.eos_token_id + 2
    )
    truncated_string = stop_at_stop_token(tokenizer.decode(generated_ids[0]), problem)
    filename = problem["name"] + "." + LANG
    with open(filename, "w") as f:
        print(f"Created {filename}")
        f.write(truncated_string)
        f.write("\n")
        f.write(problem["tests"])