Update app.py

app.py

@@ -185,16 +185,12 @@ def load_model_and_vocabs():
     with open('latin_stoi.json', 'r', encoding='utf-8') as f:
         latin_stoi = json.load(f)
     with open('latin_itos.json', 'r', encoding='utf-8') as f:
-        latin_itos_raw = json.load(f)
-        # Convert string keys to integers
-        latin_itos = {int(k): v for k, v in latin_itos_raw.items()}
+        latin_itos = json.load(f)
     
     with open('arabic_stoi.json', 'r', encoding='utf-8') as f:
         arabic_stoi = json.load(f)
     with open('arabic_itos.json', 'r', encoding='utf-8') as f:
-        arabic_itos_raw = json.load(f)
-        # Convert string keys to integers
-        arabic_itos = {int(k): v for k, v in arabic_itos_raw.items()}
+        arabic_itos= json.load(f)
     
     # Initialize model
     model = CTCTransliterator(
@@ -229,17 +225,17 @@ def greedy_decode(log_probs, blank_id, itos, stoi):
     preds = log_probs.argmax(2).T.cpu().numpy()  # (B, T)
     results = []
     raw_results = []
-
+    print(eos_id, blank_id)
     for i, pred in enumerate(preds):
         prev = None
         decoded = []
         raw_result = []
 
         for p in pred:
-            if p == eos_id:  # Stop at EOS!
+            if str(p) == eos_id:  # Stop at EOS!
                 break
             # CTC collapse: skip blanks and repeated characters
-            if p != blank_id and p != prev:
+            if str(p) != blank_id and p != prev:
                 decoded.append(itos[str(p)])  # Convert to string if needed
             prev = p
             raw_result.append(itos[str(p)])