Add SQL-BLEU metric

README.md

@@ -1,10 +1,17 @@
 ---
-title: Text2sql Sql Bleu
-emoji: 🔥
-colorFrom: yellow
-colorTo: purple
+title: text2sql-sql-bleu
+emoji: "📏"
+colorFrom: green
+colorTo: blue
 sdk: static
 pinned: false
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+Hugging Face Evaluate metric for **SQL BLEU** (token BLEU in [0,1]).
+
+Usage:
+
+```python
+import evaluate
+m = evaluate.load("3zizo3/text2sql-sql-bleu")
+print(m.compute(predictions=["select 1"], references=["select 2"]))

requirements.txt

@@ -0,0 +1,3 @@
+evaluate>=0.4.6
+datasets>=2.0.0
+git+https://github.com/AzizHamad03/text2sql-eval.git

text2sql-sql-bleu.py

@@ -0,0 +1,34 @@
+import evaluate
+import datasets
+
+from text2sql_eval.metrics.bleu import bleu_score
+
+
+_DESCRIPTION = "SQL token BLEU (0–1). Returns mean score in [0, 1]."
+
+
+def _to_str(x):
+    if isinstance(x, (list, tuple)):
+        return x[0] if x else ""
+    return "" if x is None else str(x)
+
+
+class SQLBLEU(evaluate.Metric):
+    def _info(self):
+        return evaluate.MetricInfo(
+            description=_DESCRIPTION,
+            citation="Uses sacrebleu via text2sql-eval implementation.",
+            features=datasets.Features(
+                {
+                    "predictions": datasets.Value("string"),
+                    "references": datasets.Value("string"),
+                }
+            ),
+        )
+
+    def _compute(self, predictions, references):
+        scores = []
+        for p, r in zip(predictions, references):
+            scores.append(float(bleu_score(_to_str(p), _to_str(r))))
+        mean = sum(scores) / len(scores) if scores else 0.0
+        return {"sql_bleu": mean}