test.py

import datasets
import transformers


dataset = datasets.load_dataset("syntaxgym", "mvrr_mod")
metric = datasets.load_metric("syntaxgym", "mvrr_mod")

model = transformers.AutoModelForCausalLM.from_pretrained("gpt2")


for item in dataset["test"]:
    # TODO evaluate surprisals

    print(item)

    break