Fix Nemotron-PII spans parsing with ast.literal_eval fallback

train.py

@@ -1,5 +1,5 @@
 #!/usr/bin/env python3
-import json, random, argparse
+import json, random, argparse, ast
 import numpy as np, torch
 from datasets import load_dataset, Dataset
 from transformers import AutoTokenizer, AutoModelForTokenClassification, TrainingArguments, Trainer, DataCollatorForTokenClassification
@@ -66,7 +66,10 @@ def load_nemotron_split(split, max_examples=10000):
         text = ex["text"]
         spans_raw = ex["spans"]
         if isinstance(spans_raw, str):
-            spans_raw = json.loads(spans_raw)
+            try:
+                spans_raw = json.loads(spans_raw)
+            except json.JSONDecodeError:
+                spans_raw = ast.literal_eval(spans_raw)
         spans = []
         for sp in spans_raw:
             lab = NEMOTRON_MAP.get(sp["label"])