best-of-N exceed token issue

app.py

@@ -161,7 +161,10 @@ def plan2align_translate_text(text, session_id, model, tokenizer, device, src_la
         reward_model_type=reward_model_type,
         session_id=session_id
     )
-    _, score = evaluate_candidates(text, [result], task_language, session_id)
+    try:
+        _, score = evaluate_candidates(text, [result], task_language, session_id)
+    except:
+        score = 0
     return result, score
 
 def evaluate_candidates(source, candidates, language, session_id):
@@ -178,21 +181,25 @@ def original_translation(text, src_language, target_language, session_id):
     return "", 0
 
 def best_of_n_translation(text, src_language, target_language, n, session_id):
-    if not check_token_length(text, 2048):
-         return "Warning: Input text exceeds 2048 tokens.", None, ""
+    if not check_token_length(text, 4096):
+         return "Warning: Input text too long.", 0
     candidates = []
     for i in range(n):
         cand_list = basic_translate(text, src_language, target_language)
         if cand_list:
             candidates.append(cand_list[0])
-    best, score = evaluate_candidates(text, candidates, target_language, session_id)
-    print("best_of_n evaluate_candidates results:")
-    print(best, score)
+    try:
+        best, score = evaluate_candidates(text, candidates, target_language, session_id)
+        print("best_of_n evaluate_candidates results:")
+        print(best, score)
+    except:
+        print("evaluate_candidates fail")
+        return "Warning: Input text too long.", 0
     return best, score
 
 def mpc_translation(text, src_language, target_language, iterations, session_id):
-    if not check_token_length(text, 2048):
-         return "Warning: Input text exceeds 2048 tokens.", None, ""
+    if not check_token_length(text, 4096):
+         return "Warning: Input text too long.", 0
     current_trans = ""
     best_score = None
     for i in range(iterations):
@@ -201,11 +208,17 @@ def mpc_translation(text, src_language, target_language, iterations, session_id)
         else:
             cand_list = mpc_improved_translate(text, current_trans, src_language, target_language)        
         
-        best, score = evaluate_candidates(text, cand_list, target_language, session_id)
-        print("mpc evaluate_candidates results:")
-        print(best, score)
-        current_trans = best
-        best_score = score
+        try:
+            best, score = evaluate_candidates(text, cand_list, target_language, session_id)
+            print("mpc evaluate_candidates results:")
+            print(best, score)
+            current_trans = best
+            best_score = score
+        except:
+            print("evaluate_candidates fail")
+            current_trans = cand_list[0]
+            best_score = 0
+
     return current_trans, best_score
 
 # ---------- Gradio function ----------
@@ -240,8 +253,7 @@ def process_text(text, src_language, target_language, max_iterations_value, thre
         )
         plan2align_output = f"{plan2align_trans}\n\nScore: {best_score:.2f}"
     if "Best-of-N" in translation_methods:
-        best_candidate, best_score = best_of_n_translation(text, src_language, target_language,
-                                                            max_iterations_value, session_id)
+        best_candidate, best_score = best_of_n_translation(text, src_language, target_language, max_iterations_value, session_id)
         best_of_n_output = f"{best_candidate}\n\nScore: {best_score:.2f}"
     if "MPC" in translation_methods:
         mpc_candidate, mpc_score = mpc_translation(text, src_language, target_language,