whisper-large-v2-turbo

examples/industrial_data_pretraining/whisper/demo_from_openai.py

@@ -11,9 +11,9 @@ from funasr import AutoModel
 # model = AutoModel(model="Whisper-medium", hub="openai")
 # model = AutoModel(model="Whisper-large-v2", hub="openai")
 model = AutoModel(
-    model="Whisper-large-v3",
-    vad_model="iic/speech_fsmn_vad_zh-cn-16k-common-pytorch",
-    vad_kwargs={"max_single_segment_time": 30000},
+    model="Whisper-large-v3-turbo",
+    # vad_model="iic/speech_fsmn_vad_zh-cn-16k-common-pytorch",
+    # vad_kwargs={"max_single_segment_time": 30000},
     hub="openai",
 )
 

funasr/download/name_maps_from_hub.py

@@ -47,8 +47,9 @@ name_maps_openai = {
     "Whisper-small": "small",
     "Whisper-medium.en": "medium.en",
     "Whisper-medium": "medium",
+    "Whisper-large": "large",
     "Whisper-large-v1": "large-v1",
     "Whisper-large-v2": "large-v2",
     "Whisper-large-v3": "large-v3",
-    "Whisper-large": "large",
+    "Whisper-large-v3-turbo": "turbo",
 }

funasr/download/upload_model.py

@@ -1,8 +1,25 @@
 import os
 import hydra
 from omegaconf import DictConfig, OmegaConf, ListConfig
-
+import traceback
 from modelscope.hub.api import HubApi
+from modelscope.hub.constants import (
+    API_HTTP_CLIENT_TIMEOUT,
+    API_RESPONSE_FIELD_DATA,
+    API_RESPONSE_FIELD_EMAIL,
+    API_RESPONSE_FIELD_GIT_ACCESS_TOKEN,
+    API_RESPONSE_FIELD_MESSAGE,
+    API_RESPONSE_FIELD_USERNAME,
+    DEFAULT_CREDENTIALS_PATH,
+    MODELSCOPE_CLOUD_ENVIRONMENT,
+    MODELSCOPE_CLOUD_USERNAME,
+    MODELSCOPE_REQUEST_ID,
+    ONE_YEAR_SECONDS,
+    REQUESTS_API_HTTP_METHOD,
+    DatasetVisibility,
+    Licenses,
+    ModelVisibility,
+)
 
 
 @hydra.main(config_name=None, version_base=None)
@@ -22,11 +39,41 @@ def main_hydra(cfg: DictConfig):
 
     assert TOKEN is not None, f"{TOKEN} is None"
 
+    visibility = cfg.get("visibility", "PRIVATE")  # PRIVATE #ModelVisibility.PUBLIC
+    visibility = getattr(ModelVisibility, visibility)
+    model_name = cfg.get("model_name", "测试")
+
     api = HubApi()
     api.login(TOKEN)
 
-    api.push_model(model_id=model_id, model_dir=model_dir)
+    try:
+        api.create_model(
+            model_id=model_id,
+            visibility=visibility,
+            license=Licenses.APACHE_V2,
+            chinese_name=model_name,
+        )
+    except Exception as e:
+        print(f"Create_model failed! {str(e)}, {traceback.format_exc()}")
 
+    print(f"model url: https://modelscope.cn/models/{model_id}")
+
+    api.push_model(model_id=model_id, model_dir=model_dir)
+    print(
+        f"Upload model finished."
+        f"model_dir: {model_dir}"
+        f"model_id: {model_id}"
+        f"url: https://modelscope.cn/models/{model_id}"
+    )
+
+
+"""
+TOKEN="fadd1abb-4df6-4807-9051-5ab01ac81071"
+model_id="iic/Whisper-large-v3-turbo"
+model_dir="/Users/zhifu/Downloads/Whisper-large-v3-turbo"
+
+python -m funasr.download.upload_model ++TOKEN=${TOKEN} ++model_id=${model_id} ++model_dir=${model_dir}
+"""
 
 if __name__ == "__main__":
     main_hydra()

funasr/models/whisper/model.py

@@ -28,6 +28,7 @@ from funasr.register import tables
 @tables.register("model_classes", "Whisper-large-v1")
 @tables.register("model_classes", "Whisper-large-v2")
 @tables.register("model_classes", "Whisper-large-v3")
+@tables.register("model_classes", "Whisper-large-v3-turbo")
 @tables.register("model_classes", "WhisperWarp")
 class WhisperWarp(nn.Module):
     def __init__(self, *args, **kwargs):