Skip to content

Open-source speech recognition toolkit for training, inference, streaming ASR, VAD, punctuation, speaker diarization pipelines, and OpenAI-compatible/MCP serving.

fix: remove repetition_penalty from vLLM services (fixes #2950 CUDA assert)

游雁b8d736ef8496
diff --git a/examples/industrial_data_pretraining/fun_asr_nano/serve_realtime_ws.py b/examples/industrial_data_pretraining/fun_asr_nano/serve_realtime_ws.py
index b52a1dd1d..5c58bc5ab 100644
--- a/examples/industrial_data_pretraining/fun_asr_nano/serve_realtime_ws.py
+++ b/examples/industrial_data_pretraining/fun_asr_nano/serve_realtime_ws.py
@@ -318,7 +318,6 @@ class RealtimeASRSession:
hotwords=self.asr_kwargs.get("hotwords"),
language=self.asr_kwargs.get("language"),
max_new_tokens=200,
- repetition_penalty=1.3,
)
text = results[0]["text"] if results else ""
text = _clean_asr_text(text)
@@ -359,7 +358,6 @@ class RealtimeASRSession:
hotwords=self.asr_kwargs.get("hotwords"),
language=self.asr_kwargs.get("language"),
max_new_tokens=512,
- repetition_penalty=1.3,
)
text = results[0]["text"] if results else ""
text = _clean_asr_text(text)
diff --git a/examples/industrial_data_pretraining/fun_asr_nano/serve_vllm.py b/examples/industrial_data_pretraining/fun_asr_nano/serve_vllm.py
index 1820f9517..6a0ab1a95 100644
--- a/examples/industrial_data_pretraining/fun_asr_nano/serve_vllm.py
+++ b/examples/industrial_data_pretraining/fun_asr_nano/serve_vllm.py
@@ -32,6 +32,20 @@ warnings.filterwarnings('ignore')
logging.basicConfig(level=logging.INFO, format='%(asctime)s [%(levelname)s] %(message)s')
logger = logging.getLogger(__name__)
+def truncate_repetition(text, min_repeat_len=3, max_repeats=3):
+ """Detect and truncate repetitive patterns in ASR output."""
+ if not text or len(text) < 20:
+ return text
+ n = len(text)
+ for length in range(min_repeat_len, min(n // max_repeats, 30)):
+ for start in range(n - length * max_repeats):
+ chunk = text[start:start + length]
+ if text[start:start + length * max_repeats] == chunk * max_repeats:
+ return text[:start + length]
+ return text
+
+
+
try:
from fastapi import FastAPI, File, UploadFile, Form, WebSocket, WebSocketDisconnect
from fastapi.responses import JSONResponse
@@ -107,7 +121,7 @@ def process_audio(audio_data, sr=16000, language=None, hotwords=None,
return {"text": "", "segments": [], "duration": len(audio_data) / sr}
# vLLM batch ASR
- gen_kwargs = {"max_new_tokens": 500, "repetition_penalty": 1.3}
+ gen_kwargs = {"max_new_tokens": 500}
if language:
gen_kwargs["language"] = language
if hotwords:
@@ -120,6 +134,7 @@ def process_audio(audio_data, sr=16000, language=None, hotwords=None,
full_text_parts = []
for i, (r, (start_ms, end_ms)) in enumerate(zip(results, seg_times)):
+ r["text"] = truncate_repetition(r["text"])
seg_info = {
"text": r["text"],
"start": start_ms / 1000,