minor

Signed-off-by: NickLucche <[email protected]>

Author: NickLucche

examples/online_serving/openai_transcription_client.py

@@ -3,7 +3,7 @@
 import json
 
 import httpx
-from openai import AsyncOpenAI
+from openai import OpenAI
 
 from vllm.assets.audio import AudioAsset
 
@@ -13,15 +13,15 @@
 # Modify OpenAI's API key and API base to use vLLM's API server.
 openai_api_key = "EMPTY"
 openai_api_base = "http://localhost:8000/v1"
-client = AsyncOpenAI(
+client = OpenAI(
     api_key=openai_api_key,
     base_url=openai_api_base,
 )
 
 
-async def main():
+def sync_openai():
     with open(str(mary_had_lamb), "rb") as f:
-        transcription = await client.audio.transcriptions.create(
+        transcription = client.audio.transcriptions.create(
             file=f,
             model="openai/whisper-small",
             language="en",
@@ -30,7 +30,7 @@ async def main():
         print("transcription result:", transcription.text)
 
 
-asyncio.run(main())
+sync_openai()
 
 
 # OpenAI Transcription API client does not support streaming.

vllm/entrypoints/openai/serving_transcription.py

@@ -346,8 +346,8 @@ async def transcription_stream_generator(
                     # NOTE(NickLucche) user can't pass encoder prompts directly
                     # at least not to Whisper. One indicator of the encoder
                     # amount of processing is the log-mel spectogram length.
-                    num_prompt_tokens = ceil(audio_duration_s * self.model_sr /
-                                             self.hop_length)
+                    num_prompt_tokens += ceil(audio_duration_s *
+                                              self.model_sr / self.hop_length)
 
                 # We need to do it here, because if there are exceptions in
                 # the result_generator, it needs to be sent as the FIRST