Commit @4aa5cf9c49b0becfb5282f0e8e0bb4d48c00c9e5 - yjyoon/whisper_server

Fedir Zadniprovskyi 01-13

deprecate `config.preload_models`

@4aa5cf9c49b0becfb5282f0e8e0bb4d48c00c9e5

8f9d361

4aa5cf9

src/speaches/config.py

--- src/speaches/config.py

+++ src/speaches/config.py


     """
     default_response_format: ResponseFormat = ResponseFormat.JSON
     whisper: WhisperConfig = WhisperConfig()
-    preload_models: list[str] = Field(
-        default_factory=list,
-        examples=[
-            ["Systran/faster-whisper-small"],
-            ["Systran/faster-whisper-medium.en", "Systran/faster-whisper-small.en"],
-        ],
-    )
-    """
-    List of Whisper models to preload on startup. By default, the model is first loaded on first request.
-    WARNING: I'd recommend not setting this, as it may be deprecated in the future.
-    """
     max_no_data_seconds: float = 1.0
     """
     Max duration to wait for the next audio chunk before transcription is finilized and connection is closed.

8f9d361

4aa5cf9

src/speaches/main.py

--- src/speaches/main.py

+++ src/speaches/main.py


 from __future__ import annotations
 
-from contextlib import asynccontextmanager
 import logging
 import platform
-from typing import TYPE_CHECKING
 
 from fastapi import (
     FastAPI,
 )
 from fastapi.middleware.cors import CORSMiddleware
 
-from speaches.dependencies import ApiKeyDependency, get_config, get_model_manager
+from speaches.dependencies import ApiKeyDependency, get_config
 from speaches.logger import setup_logger
 from speaches.routers.misc import (
     router as misc_router,

 from speaches.routers.stt import (
     router as stt_router,
 )
-
-if TYPE_CHECKING:
-    from collections.abc import AsyncGenerator
 
 # https://swagger.io/docs/specification/v3_0/grouping-operations-with-tags/
 # https://fastapi.tiangolo.com/tutorial/metadata/#metadata-for-tags

     if platform.machine() == "x86_64":
         logger.warning("`POST /v1/audio/speech` with `model=rhasspy/piper-voices` is only supported on x86_64 machines")
 
-    model_manager = get_model_manager()  # HACK
-
-    @asynccontextmanager
-    async def lifespan(_app: FastAPI) -> AsyncGenerator[None, None]:
-        for model_name in config.preload_models:
-            model_manager.load_model(model_name)
-        yield
-
     dependencies = []
     if config.api_key is not None:
         dependencies.append(ApiKeyDependency)
 
-    app = FastAPI(lifespan=lifespan, dependencies=dependencies, openapi_tags=TAGS_METADATA)
+    app = FastAPI(dependencies=dependencies, openapi_tags=TAGS_METADATA)
 
     app.include_router(stt_router)
     app.include_router(models_router)

Add a comment

Open 0
Closed 0

List

...	...	@@ -216,17 +216,6 @@
216	216	"""
217	217	default_response_format: ResponseFormat = ResponseFormat.JSON
218	218	whisper: WhisperConfig = WhisperConfig()
219		- preload_models: list[str] = Field(
220		- default_factory=list,
221		- examples=[
222		- ["Systran/faster-whisper-small"],
223		- ["Systran/faster-whisper-medium.en", "Systran/faster-whisper-small.en"],
224		- ],
225		- )
226		- """
227		- List of Whisper models to preload on startup. By default, the model is first loaded on first request.
228		- WARNING: I'd recommend not setting this, as it may be deprecated in the future.
229		- """
230	219	max_no_data_seconds: float = 1.0
231	220	"""
232	221	Max duration to wait for the next audio chunk before transcription is finilized and connection is closed.

...	...	@@ -1,16 +1,14 @@
1	1	from __future__ import annotations
2	2
3		-from contextlib import asynccontextmanager
4	3	import logging
5	4	import platform
6		-from typing import TYPE_CHECKING
7	5
8	6	from fastapi import (
9	7	FastAPI,
10	8	)
11	9	from fastapi.middleware.cors import CORSMiddleware
12	10
13		-from speaches.dependencies import ApiKeyDependency, get_config, get_model_manager
	11	+from speaches.dependencies import ApiKeyDependency, get_config
14	12	from speaches.logger import setup_logger
15	13	from speaches.routers.misc import (
16	14	router as misc_router,
...	...	@@ -24,9 +22,6 @@
24	22	from speaches.routers.stt import (
25	23	router as stt_router,
26	24	)
27		-
28		-if TYPE_CHECKING:
29		- from collections.abc import AsyncGenerator
30	25
31	26	# https://swagger.io/docs/specification/v3_0/grouping-operations-with-tags/
32	27	# https://fastapi.tiangolo.com/tutorial/metadata/#metadata-for-tags
...	...	@@ -52,19 +47,11 @@
52	47	if platform.machine() == "x86_64":
53	48	logger.warning("`POST /v1/audio/speech` with `model=rhasspy/piper-voices` is only supported on x86_64 machines")
54	49
55		- model_manager = get_model_manager() # HACK
56		-
57		- @asynccontextmanager
58		- async def lifespan(_app: FastAPI) -> AsyncGenerator[None, None]:
59		- for model_name in config.preload_models:
60		- model_manager.load_model(model_name)
61		- yield
62		-
63	50	dependencies = []
64	51	if config.api_key is not None:
65	52	dependencies.append(ApiKeyDependency)
66	53
67		- app = FastAPI(lifespan=lifespan, dependencies=dependencies, openapi_tags=TAGS_METADATA)
	54	+ app = FastAPI(dependencies=dependencies, openapi_tags=TAGS_METADATA)
68	55
69	56	app.include_router(stt_router)
70	57	app.include_router(models_router)

Delete comment