src/llmtuner/train/sft/metric.py

from dataclasses import dataclass
from typing import TYPE_CHECKING, Dict, Sequence, Tuple, Union

import numpy as np

from ...extras.constants import IGNORE_INDEX
from ...extras.packages import is_jieba_available, is_nltk_available, is_rouge_available


if TYPE_CHECKING:
    from transformers.tokenization_utils import PreTrainedTokenizer

if is_jieba_available():
    import jieba

if is_nltk_available():
    from nltk.translate.bleu_score import SmoothingFunction, sentence_bleu

if is_rouge_available():
    from rouge_chinese import Rouge


@dataclass
class ComputeMetrics:
    r"""
    Wraps the tokenizer into metric functions, used in Seq2SeqPeftTrainer.
    """

    tokenizer: "PreTrainedTokenizer"

    def __call__(self, eval_preds: Sequence[Union[np.ndarray, Tuple[np.ndarray]]]) -> Dict[str, float]:
        r"""
        Uses the model predictions to compute metrics.
        """
        preds, labels = eval_preds
        score_dict = {"rouge-1": [], "rouge-2": [], "rouge-l": [], "bleu-4": []}

        preds = np.where(preds != IGNORE_INDEX, preds, self.tokenizer.pad_token_id)
        labels = np.where(labels != IGNORE_INDEX, labels, self.tokenizer.pad_token_id)

        decoded_preds = self.tokenizer.batch_decode(preds, skip_special_tokens=True)
        decoded_labels = self.tokenizer.batch_decode(labels, skip_special_tokens=True)

        for pred, label in zip(decoded_preds, decoded_labels):
            hypothesis = list(jieba.cut(pred))
            reference = list(jieba.cut(label))

            if len(" ".join(hypothesis).split()) == 0 or len(" ".join(reference).split()) == 0:
                result = {"rouge-1": {"f": 0.0}, "rouge-2": {"f": 0.0}, "rouge-l": {"f": 0.0}}
            else:
                rouge = Rouge()
                scores = rouge.get_scores(" ".join(hypothesis), " ".join(reference))
                result = scores[0]

            for k, v in result.items():
                score_dict[k].append(round(v["f"] * 100, 4))

            bleu_score = sentence_bleu([list(label)], list(pred), smoothing_function=SmoothingFunction().method3)
            score_dict["bleu-4"].append(round(bleu_score * 100, 4))

        return {k: float(np.mean(v)) for k, v in score_dict.items()}