Commit @3dda14e4a25775cdba0188cdc4f656e2248c86d0 - yjyoon/whisper_server

Fedir Zadniprovskyi 2024-06-13

chore: handle "whisper-1" model name

@3dda14e4a25775cdba0188cdc4f656e2248c86d0

f89c747

3dda14e

faster_whisper_server/main.py

--- faster_whisper_server/main.py

+++ faster_whisper_server/main.py


     FastAPI,
     Form,
     HTTPException,
+    Path,
     Query,
     Response,
     UploadFile,

 from faster_whisper import WhisperModel
 from faster_whisper.vad import VadOptions, get_speech_timestamps
 from huggingface_hub.hf_api import ModelInfo
+from pydantic import AfterValidator
 
 from faster_whisper_server import utils
 from faster_whisper_server.asr import FasterWhisperASR

     return Response(status_code=200, content="OK")
 
 
-@app.get("/v1/models", response_model=list[ModelObject])
+@app.get("/v1/models")
 def get_models() -> list[ModelObject]:
     models = huggingface_hub.list_models(library="ctranslate2")
     models = [

     return models
 
 
-@app.get("/v1/models/{model_name:path}", response_model=ModelObject)
-def get_model(model_name: str) -> ModelObject:
+@app.get("/v1/models/{model_name:path}")
+def get_model(model_name: Annotated[str, Path()]) -> ModelObject:
     models = list(
         huggingface_hub.list_models(model_name=model_name, library="ctranslate2")
     )

     return f"data: {data}\n\n"
 
 
+def handle_default_openai_model(model_name: str) -> str:
+    """This exists because some callers may not be able override the default("whisper-1") model name.
+    For example, https://github.com/open-webui/open-webui/issues/2248#issuecomment-2162997623.
+    """
+    if model_name == "whisper-1":
+        logger.info(
+            f"{model_name} is not a valid model name. Using {config.whisper.model} instead."
+        )
+        return config.whisper.model
+    return model_name
+
+
+ModelName = Annotated[str, AfterValidator(handle_default_openai_model)]
+
+
 @app.post("/v1/audio/translations")
 def translate_file(
     file: Annotated[UploadFile, Form()],
-    model: Annotated[str, Form()] = config.whisper.model,
+    model: Annotated[ModelName, Form()] = config.whisper.model,
     prompt: Annotated[str | None, Form()] = None,
     response_format: Annotated[ResponseFormat, Form()] = config.default_response_format,
     temperature: Annotated[float, Form()] = 0.0,

 @app.post("/v1/audio/transcriptions")
 def transcribe_file(
     file: Annotated[UploadFile, Form()],
-    model: Annotated[str, Form()] = config.whisper.model,
+    model: Annotated[ModelName, Form()] = config.whisper.model,
     language: Annotated[Language | None, Form()] = config.default_language,
     prompt: Annotated[str | None, Form()] = None,
     response_format: Annotated[ResponseFormat, Form()] = config.default_response_format,

 @app.websocket("/v1/audio/transcriptions")
 async def transcribe_stream(
     ws: WebSocket,
-    model: Annotated[str, Query()] = config.whisper.model,
+    model: Annotated[ModelName, Query()] = config.whisper.model,
     language: Annotated[Language | None, Query()] = config.default_language,
     response_format: Annotated[
         ResponseFormat, Query()

Add a comment

Open 0
Closed 0

List

...	...	@@ -11,6 +11,7 @@
11	11	FastAPI,
12	12	Form,
13	13	HTTPException,
	14	+ Path,
14	15	Query,
15	16	Response,
16	17	UploadFile,
...	...	@@ -22,6 +23,7 @@
22	23	from faster_whisper import WhisperModel
23	24	from faster_whisper.vad import VadOptions, get_speech_timestamps
24	25	from huggingface_hub.hf_api import ModelInfo
	26	+from pydantic import AfterValidator
25	27
26	28	from faster_whisper_server import utils
27	29	from faster_whisper_server.asr import FasterWhisperASR
...	...	@@ -85,7 +87,7 @@
85	87	return Response(status_code=200, content="OK")
86	88
87	89
88		-@app.get("/v1/models", response_model=list[ModelObject])
	90	+@app.get("/v1/models")
89	91	def get_models() -> list[ModelObject]:
90	92	models = huggingface_hub.list_models(library="ctranslate2")
91	93	models = [
...	...	@@ -101,8 +103,8 @@
101	103	return models
102	104
103	105
104		-@app.get("/v1/models/{model_name:path}", response_model=ModelObject)
105		-def get_model(model_name: str) -> ModelObject:
	106	+@app.get("/v1/models/{model_name:path}")
	107	+def get_model(model_name: Annotated[str, Path()]) -> ModelObject:
106	108	models = list(
107	109	huggingface_hub.list_models(model_name=model_name, library="ctranslate2")
108	110	)
...	...	@@ -131,10 +133,25 @@
131	133	return f"data: {data}\n\n"
132	134
133	135
	136	+def handle_default_openai_model(model_name: str) -> str:
	137	+ """This exists because some callers may not be able override the default("whisper-1") model name.
	138	+ For example, https://github.com/open-webui/open-webui/issues/2248#issuecomment-2162997623.
	139	+ """
	140	+ if model_name == "whisper-1":
	141	+ logger.info(
	142	+ f"{model_name} is not a valid model name. Using {config.whisper.model} instead."
	143	+ )
	144	+ return config.whisper.model
	145	+ return model_name
	146	+
	147	+
	148	+ModelName = Annotated[str, AfterValidator(handle_default_openai_model)]
	149	+
	150	+
134	151	@app.post("/v1/audio/translations")
135	152	def translate_file(
136	153	file: Annotated[UploadFile, Form()],
137		- model: Annotated[str, Form()] = config.whisper.model,
	154	+ model: Annotated[ModelName, Form()] = config.whisper.model,
138	155	prompt: Annotated[str \| None, Form()] = None,
139	156	response_format: Annotated[ResponseFormat, Form()] = config.default_response_format,
140	157	temperature: Annotated[float, Form()] = 0.0,
...	...	@@ -187,7 +204,7 @@
187	204	@app.post("/v1/audio/transcriptions")
188	205	def transcribe_file(
189	206	file: Annotated[UploadFile, Form()],
190		- model: Annotated[str, Form()] = config.whisper.model,
	207	+ model: Annotated[ModelName, Form()] = config.whisper.model,
191	208	language: Annotated[Language \| None, Form()] = config.default_language,
192	209	prompt: Annotated[str \| None, Form()] = None,
193	210	response_format: Annotated[ResponseFormat, Form()] = config.default_response_format,
...	...	@@ -289,7 +306,7 @@
289	306	@app.websocket("/v1/audio/transcriptions")
290	307	async def transcribe_stream(
291	308	ws: WebSocket,
292		- model: Annotated[str, Query()] = config.whisper.model,
	309	+ model: Annotated[ModelName, Query()] = config.whisper.model,
293	310	language: Annotated[Language \| None, Query()] = config.default_language,
294	311	response_format: Annotated[
295	312	ResponseFormat, Query()

Delete comment