utils.py

# noqa
import itertools
import re
from functools import cache
from typing import Literal, Generator, Union, TYPE_CHECKING

import datasets
from transformers import AutoTokenizer

from lm_eval.tasks.ruler.essays import get_all_essays
from lm_eval.tasks.ruler.prepare import generate_samples

if TYPE_CHECKING:
    import transformers


def get_tokenizer(
    **kwargs,
) -> Union["transformers.PreTrainedTokenizer", "transformers.PreTrainedTokenizerFast"]:
    kwargs = kwargs.get("metadata", {})
    pretrained = kwargs.get("tokenizer", kwargs.get("pretrained", {}))
    assert pretrained, "No tokenizer or pretrained provided."
    print("using tokenizer ", pretrained)
    return AutoTokenizer.from_pretrained(pretrained, trust_remote_code=True)


# TOKENIZER = AutoTokenizer.from_pretrained(os.environ.get("TOKENIZER"))
TEMPLATE = """Some special magic {type_needle_v} are hidden within the following text. Make sure to memorize it. I will quiz you about the {type_needle_v} afterwards.\n{context}\nWhat are all the special magic {type_needle_v} for {query} mentioned in the provided text? The special magic {type_needle_v} for {query} mentioned in the provided text are"""

SEQ_LENGTHS = (
    # 131072,
    # 65536,
    # 32768,
    16384,
    8192,
    4096,
)

NUM_SAMPLES = 500
REMOVE_NEWLINE_TAB = ""
STOP_WORDS = ""
RANDOM_SEED = 42


@cache
def get_haystack(
    type_haystack: Literal["essay", "repeat", "needle"],
) -> Union[list[str], str]:
    NEEDLE = "One of the special magic {type_needle_v} for {key} is: {value}."
    if type_haystack == "essay":
        essay = get_all_essays()["text"]
        # essay = json.load(open(essay))["text"]
        haystack = re.sub(r"\s+", " ", essay).split(" ")
    elif type_haystack == "repeat":
        haystack = "The grass is green. The sky is blue. The sun is yellow. Here we go. There and back again."
    elif type_haystack == "needle":
        haystack = NEEDLE
    else:
        raise NotImplementedError(f"{type_haystack} is not implemented.")
    return haystack


def flatten(df: Generator) -> dict[str, datasets.Dataset]:
    return {
        "test": datasets.Dataset.from_list(
            list(itertools.chain.from_iterable(df)), split=datasets.Split.TEST
        )
    }


# ruff: noqa
niah_single_1 = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="repeat"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="repeat",
        type_needle_k="words",
        type_needle_v="numbers",
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)
# ruff: noqa
niah_single_2 = lambda x: flatten(
    generate_samples(
        get_haystack(type_haystack="essay"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="essay",
        type_needle_k="words",
        type_needle_v="numbers",
        TOKENIZER=get_tokenizer(x),
    )
    for seq in SEQ_LENGTHS
)
# noqa
niah_single_3 = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="essay"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="essay",
        type_needle_k="words",
        type_needle_v="uuids",
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)
# noqa
niah_multikey_1 = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="essay"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="essay",
        type_needle_k="words",
        type_needle_v="numbers",
        num_needle_k=4,
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)
# noqa
niah_multikey_2 = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="needle"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="needle",
        type_needle_k="words",
        type_needle_v="numbers",
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)
# noqa
niah_multikey_3 = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="needle"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="needle",
        type_needle_k="uuids",
        type_needle_v="uuids",
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)
# noqa
niah_multivalue = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="essay"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="essay",
        type_needle_k="words",
        type_needle_v="numbers",
        num_needle_v=4,
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)
# noqa
niah_multiquery = lambda **kwargs: flatten(
    generate_samples(
        get_haystack(type_haystack="essay"),
        max_seq_length=seq,
        template=TEMPLATE,
        type_haystack="essay",
        type_needle_k="words",
        type_needle_v="numbers",
        num_needle_q=4,
        TOKENIZER=get_tokenizer(**kwargs),
    )
    for seq in SEQ_LENGTHS
)


def postprocess_pred(predict_str: str) -> str:
    predict_str = predict_str.strip()

    # Remove all non-printable characters
    np_pattern = re.compile(r"[\x00-\x1f]")
    predict_str = np_pattern.sub("\n", predict_str).strip()

    return predict_str


def string_match_all(preds: list[str], refs: list[list[str]]) -> float:
    score = sum(
        [
            sum([1.0 if r.lower() in pred.lower() else 0.0 for r in ref]) / len(ref)
            for pred, ref in zip(preds, refs)
        ]
    ) / len(preds)
    return score


def process_results(doc: dict, results: list[str]) -> dict[str, float]:
    # hacky: set all other lengths to -1
    metrics = {str(length): -1.0 for length in SEQ_LENGTHS}
    input_len = doc["max_length"]
    pred = postprocess_pred(results[0])
    score = string_match_all([pred], [doc["outputs"]])
    metrics[str(input_len)] = score
    return metrics


def aggregate_metrics(metrics: list[float]) -> float:
    res = [x for x in metrics if x != -1]
    if not res:
        # we don't have any samples with this length
        return 0.0
    return sum(res) / len(res)