SoftALL/text2sql-sql-bleu

1import evaluate2import datasets3 4from text2sql_eval.metrics.bleu import bleu_score5 6 7_DESCRIPTION = "SQL token BLEU (0–1). Returns mean score in [0, 1]."8 9 10def _to_str(x):11    if isinstance(x, (list, tuple)):12        return x[0] if x else ""13    return "" if x is None else str(x)14 15 16class SQLBLEU(evaluate.Metric):17    def _info(self):18        return evaluate.MetricInfo(19            description=_DESCRIPTION,20            citation="Uses sacrebleu via text2sql-eval implementation.",21            features=datasets.Features(22                {23                    "predictions": datasets.Value("string"),24                    "references": datasets.Value("string"),25                }26            ),27        )28 29    def _compute(self, predictions, references):30        scores = []31        for p, r in zip(predictions, references):32            scores.append(float(bleu_score(_to_str(p), _to_str(r))))33        mean = sum(scores) / len(scores) if scores else 0.034        return {"sql_bleu": mean}35