diff --git a/backend/app/services/stt.py b/backend/app/services/stt.py
index 952de86..3c04f18 100644
--- a/backend/app/services/stt.py
+++ b/backend/app/services/stt.py
@@ -23,7 +23,7 @@ from boson_multimodal.data_collator.higgs_audio_collator import HiggsAudioSample
 from boson_multimodal.data_types import ChatMLSample, AudioContent, Message
 from boson_multimodal.dataset.chatml_dataset import (
     ChatMLDatasetSample,
-    prepare_chatml_sample_qwen,
+    prepare_chatml_sample,
 )
 
 from ..config import settings
@@ -96,8 +96,7 @@ def transcribe_segment(audio_path: str, start_time: float, end_time: float) -> s
         Message(role="user", content=[prompt, AudioContent(audio_url="placeholder")])
     ]
     chatml = ChatMLSample(messages=messages)
-    prep_fn = partial(prepare_chatml_sample_qwen, enable_thinking=True)
-    input_tokens, _, _, _ = prep_fn(chatml, _tokenizer, add_generation_prompt=True)
+    input_tokens, _, _, _ = prepare_chatml_sample(chatml, _tokenizer)
 
     sample = ChatMLDatasetSample(
         input_ids=torch.LongTensor(input_tokens),