diff --git a/src/seamless_communication/streaming/agents/online_text_decoder.py b/src/seamless_communication/streaming/agents/online_text_decoder.py
index 2bb54a75..3b5fedda 100644
--- a/src/seamless_communication/streaming/agents/online_text_decoder.py
+++ b/src/seamless_communication/streaming/agents/online_text_decoder.py
@@ -325,10 +325,7 @@ def policy(self, states: DecoderAgentStates) -> Action:
         blocked_ngrams = self.get_blocked_ngrams(states.target_indices)
         decoder_features_out = None
 
-        while (
-            len(states.target_indices + pred_indices) < self.max_len(states)
-            and len(pred_indices) < self.max_consecutive_writes
-        ):
+        while True:
             index, prob, decoder_features = self.run_decoder(states, pred_indices)
 
             if decoder_features_out is None:
@@ -361,6 +358,12 @@ def policy(self, states: DecoderAgentStates) -> Action:
             if prob < self.decision_threshold and not states.source_finished:
                 break
 
+            if (
+                len(states.target_indices + pred_indices) >= self.max_len(states)
+                or len(pred_indices) >= self.max_consecutive_writes
+            ):
+                break
+
             pred_indices.append(index)
             if self.state_bag.step_nr == 0:
                 self.state_bag.increment_step_nr(