diff --git a/turbo_alignment/dataset/chat/chat.py b/turbo_alignment/dataset/chat/chat.py
index a1409ba..b722d45 100755
--- a/turbo_alignment/dataset/chat/chat.py
+++ b/turbo_alignment/dataset/chat/chat.py
@@ -193,8 +193,8 @@ def _truncate_and_merge(
             ]
             right_bound = random.choice(bot_indices) if bot_indices else right_bound
 
-        input_ids = np.array([self.tokenizer.bos_token_id])
-        labels = np.array([DISABLE_LOSS_LABEL])
+        input_ids = np.array([])
+        labels = np.array([])
 
         truncated_conversation_messages = conversation.messages[left_bound:right_bound]
         truncated_tokenized_replicas = tokenized_replicas[left_bound:right_bound]
@@ -205,8 +205,8 @@ def _truncate_and_merge(
 
         for ind, (message, tokenized_replica) in enumerate(
             zip(
-                truncated_conversation_messages[left_bound:right_bound],
-                truncated_tokenized_replicas[left_bound:right_bound],
+                truncated_conversation_messages,
+                truncated_tokenized_replicas,
             )
         ):
             prefix_tokens = role_prefix_tokens[message.role]
@@ -236,6 +236,20 @@ def _truncate_and_merge(
             input_ids = np.append(input_ids, self.tokenizer.eos_token_id)
             labels = np.append(labels, DISABLE_LOSS_LABEL)
 
+        # FIXME
+        start_replica_token_id = role_prefix_tokens[ChatMessageRole.BOT][0].item()
+
+        # -1 for bos token
+        input_ids = input_ids[-(self.settings.max_tokens_count - 1) :]  # type: ignore[operator]
+        replica_start_token_inds = np.where(input_ids == start_replica_token_id)[0]
+        if len(replica_start_token_inds) != 0:
+            cut_index = replica_start_token_inds[0]
+            input_ids = input_ids[cut_index:]
+
+        labels = labels[-len(input_ids) :]
+        input_ids = np.concatenate((np.array([self.tokenizer.bos_token_id]), input_ids))
+        labels = np.concatenate((np.array([DISABLE_LOSS_LABEL]), labels))
+
         return input_ids, labels, conversation.get_prompt_repr(left_bound, right_bound)
 
     def _encode(