chore: import upstream snapshot with attribution

examples/eval-bert-score/bertscore_check.py

@@ -0,0 +1,26 @@
+"""
+BERTScore - measure semantic similarity between generated and reference text.
+"""
+
+import logging
+
+from bert_score import score
+
+logger = logging.getLogger(__name__)
+
+
+def get_assert(output, context):
+    """
+    Returns BERTScore F1 as a number (0-1).
+    Promptfoo will compare against threshold automatically.
+    """
+    reference = context.get("vars", {}).get("reference", "")
+    if not reference:
+        return 0
+
+    try:
+        _, _, F1 = score([output], [reference], lang="en", verbose=False)
+        return F1.item()
+    except Exception:
+        logger.debug("BERTScore calculation failed", exc_info=True)
+        return 0