Commit @844928f65c8a789975618ae061accb7260e8da9b - yjyoon/whisper_server

Fedir Zadniprovskyi 2024-07-20

refactor

@844928f65c8a789975618ae061accb7260e8da9b

905447e

844928f

faster_whisper_server/config.py

--- faster_whisper_server/config.py

+++ faster_whisper_server/config.py


     model_config = SettingsConfigDict(env_nested_delimiter="__")
 
     log_level: str = "info"
+    host: str = Field(alias="UVICORN_HOST", default="0.0.0.0")
+    port: int = Field(alias="UVICORN_PORT", default=8000)
+
     default_language: Language | None = None
     default_response_format: ResponseFormat = ResponseFormat.JSON
     whisper: WhisperConfig = WhisperConfig()

905447e

844928f

faster_whisper_server/gradio_app.py

--- faster_whisper_server/gradio_app.py

+++ faster_whisper_server/gradio_app.py


 from collections.abc import Generator
-import os
 
 import gradio as gr
 import httpx

 TRANSCRIPTION_ENDPOINT = "/v1/audio/transcriptions"
 TRANSLATION_ENDPOINT = "/v1/audio/translations"
 TIMEOUT_SECONDS = 180
+TIMEOUT = httpx.Timeout(timeout=TIMEOUT_SECONDS)
 
 
 def create_gradio_demo(config: Config) -> gr.Blocks:
-    host = os.getenv("UVICORN_HOST", "0.0.0.0")
-    port = int(os.getenv("UVICORN_PORT", "8000"))
-    # NOTE: worth looking into generated clients
-    http_client = httpx.Client(base_url=f"http://{host}:{port}", timeout=httpx.Timeout(timeout=TIMEOUT_SECONDS))
-    openai_client = OpenAI(base_url=f"http://{host}:{port}/v1", api_key="cant-be-empty")
+    base_url = f"http://{config.host}:{config.port}"
+    http_client = httpx.Client(base_url=base_url, timeout=TIMEOUT)
+    openai_client = OpenAI(base_url=f"{base_url}/v1", api_key="cant-be-empty")
 
     def handler(file_path: str, model: str, task: Task, temperature: float, stream: bool) -> Generator[str, None, None]:
-        if stream:
-            previous_transcription = ""
-            for transcription in transcribe_audio_streaming(file_path, task, temperature, model):
-                previous_transcription += transcription
-                yield previous_transcription
-        else:
-            yield transcribe_audio(file_path, task, temperature, model)
-
-    def transcribe_audio(file_path: str, task: Task, temperature: float, model: str) -> str:
         if task == Task.TRANSCRIBE:
             endpoint = TRANSCRIPTION_ENDPOINT
         elif task == Task.TRANSLATE:
             endpoint = TRANSLATION_ENDPOINT
 
+        if stream:
+            previous_transcription = ""
+            for transcription in streaming_audio_task(file_path, endpoint, temperature, model):
+                previous_transcription += transcription
+                yield previous_transcription
+        else:
+            yield audio_task(file_path, endpoint, temperature, model)
+
+    def audio_task(file_path: str, endpoint: str, temperature: float, model: str) -> str:
         with open(file_path, "rb") as file:
             response = http_client.post(
                 endpoint,

         response.raise_for_status()
         return response.text
 
-    def transcribe_audio_streaming(
-        file_path: str, task: Task, temperature: float, model: str
+    def streaming_audio_task(
+        file_path: str, endpoint: str, temperature: float, model: str
     ) -> Generator[str, None, None]:
         with open(file_path, "rb") as file:
             kwargs = {

                     "stream": True,
                 },
             }
-            endpoint = TRANSCRIPTION_ENDPOINT if task == Task.TRANSCRIBE else TRANSLATION_ENDPOINT
             with connect_sse(http_client, "POST", endpoint, **kwargs) as event_source:
                 for event in event_source.iter_sse():
                     yield event.data

Add a comment

Open 0
Closed 0

List

...	...	@@ -195,6 +195,9 @@
195	195	model_config = SettingsConfigDict(env_nested_delimiter="__")
196	196
197	197	log_level: str = "info"
	198	+ host: str = Field(alias="UVICORN_HOST", default="0.0.0.0")
	199	+ port: int = Field(alias="UVICORN_PORT", default=8000)
	200	+
198	201	default_language: Language \| None = None
199	202	default_response_format: ResponseFormat = ResponseFormat.JSON
200	203	whisper: WhisperConfig = WhisperConfig()

...	...	@@ -1,5 +1,4 @@
1	1	from collections.abc import Generator
2		-import os
3	2
4	3	import gradio as gr
5	4	import httpx
...	...	@@ -11,30 +10,29 @@
11	10	TRANSCRIPTION_ENDPOINT = "/v1/audio/transcriptions"
12	11	TRANSLATION_ENDPOINT = "/v1/audio/translations"
13	12	TIMEOUT_SECONDS = 180
	13	+TIMEOUT = httpx.Timeout(timeout=TIMEOUT_SECONDS)
14	14
15	15
16	16	def create_gradio_demo(config: Config) -> gr.Blocks:
17		- host = os.getenv("UVICORN_HOST", "0.0.0.0")
18		- port = int(os.getenv("UVICORN_PORT", "8000"))
19		- # NOTE: worth looking into generated clients
20		- http_client = httpx.Client(base_url=f"http://{host}:{port}", timeout=httpx.Timeout(timeout=TIMEOUT_SECONDS))
21		- openai_client = OpenAI(base_url=f"http://{host}:{port}/v1", api_key="cant-be-empty")
	17	+ base_url = f"http://{config.host}:{config.port}"
	18	+ http_client = httpx.Client(base_url=base_url, timeout=TIMEOUT)
	19	+ openai_client = OpenAI(base_url=f"{base_url}/v1", api_key="cant-be-empty")
22	20
23	21	def handler(file_path: str, model: str, task: Task, temperature: float, stream: bool) -> Generator[str, None, None]:
24		- if stream:
25		- previous_transcription = ""
26		- for transcription in transcribe_audio_streaming(file_path, task, temperature, model):
27		- previous_transcription += transcription
28		- yield previous_transcription
29		- else:
30		- yield transcribe_audio(file_path, task, temperature, model)
31		-
32		- def transcribe_audio(file_path: str, task: Task, temperature: float, model: str) -> str:
33	22	if task == Task.TRANSCRIBE:
34	23	endpoint = TRANSCRIPTION_ENDPOINT
35	24	elif task == Task.TRANSLATE:
36	25	endpoint = TRANSLATION_ENDPOINT
37	26
	27	+ if stream:
	28	+ previous_transcription = ""
	29	+ for transcription in streaming_audio_task(file_path, endpoint, temperature, model):
	30	+ previous_transcription += transcription
	31	+ yield previous_transcription
	32	+ else:
	33	+ yield audio_task(file_path, endpoint, temperature, model)
	34	+
	35	+ def audio_task(file_path: str, endpoint: str, temperature: float, model: str) -> str:
38	36	with open(file_path, "rb") as file:
39	37	response = http_client.post(
40	38	endpoint,
...	...	@@ -49,8 +47,8 @@
49	47	response.raise_for_status()
50	48	return response.text
51	49
52		- def transcribe_audio_streaming(
53		- file_path: str, task: Task, temperature: float, model: str
	50	+ def streaming_audio_task(
	51	+ file_path: str, endpoint: str, temperature: float, model: str
54	52	) -> Generator[str, None, None]:
55	53	with open(file_path, "rb") as file:
56	54	kwargs = {
...	...	@@ -62,7 +60,6 @@
62	60	"stream": True,
63	61	},
64	62	}
65		- endpoint = TRANSCRIPTION_ENDPOINT if task == Task.TRANSCRIBE else TRANSLATION_ENDPOINT
66	63	with connect_sse(http_client, "POST", endpoint, **kwargs) as event_source:
67	64	for event in event_source.iter_sse():
68	65	yield event.data

Delete comment