Skip to content

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

refactor: Qwen3ASR follows standard AutoModel hub download convention

游雁06e03d7baba3
diff --git a/funasr/models/qwen3_asr/model.py b/funasr/models/qwen3_asr/model.py
index 16db933e0..bf666f137 100644
--- a/funasr/models/qwen3_asr/model.py
+++ b/funasr/models/qwen3_asr/model.py
@@ -10,11 +10,13 @@ from funasr.register import tables
@tables.register("model_classes", "Qwen3ASR")
[email protected]("model_classes", "Qwen/Qwen3-ASR-1.7B")
[email protected]("model_classes", "Qwen/Qwen3-ASR-0.6B")
class Qwen3ASR(nn.Module):
def __init__(self, **kwargs):
super().__init__()
- model_path = kwargs.get("model_path", "Qwen/Qwen3-ASR-1.7B")
+ model_path = kwargs.get("model_path", kwargs.get("model", "Qwen/Qwen3-ASR-1.7B"))
device = kwargs.get("device", "cuda:0")
dtype = kwargs.get("dtype", "bf16")
max_new_tokens = kwargs.get("max_new_tokens", 512)
diff --git a/tests_models/test_qwen3_asr.py b/tests_models/test_qwen3_asr.py
index 1934294a4..1020795d7 100644
--- a/tests_models/test_qwen3_asr.py
+++ b/tests_models/test_qwen3_asr.py
@@ -7,20 +7,16 @@ import time
def main():
from funasr import AutoModel
- model_path = "Qwen/Qwen3-ASR-1.7B"
url_zh = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_zh.wav"
url_en = "https://qianwen-res.oss-cn-beijing.aliyuncs.com/Qwen3-ASR-Repo/asr_en.wav"
+ # Standard FunASR usage: model path from ModelScope/HuggingFace
+ # Default hub="ms" (ModelScope), use hub="hf" for HuggingFace
print("[Qwen3-ASR] Loading model...")
t0 = time.time()
model = AutoModel(
- model="Qwen3ASR",
- model_conf={
- "model_path": model_path,
- "dtype": "bf16",
- "max_new_tokens": 256,
- "max_inference_batch_size": 32,
- },
+ model="Qwen/Qwen3-ASR-1.7B",
+ hub="hf",
device="cuda:0",
disable_update=True,
)