[Bug] LiteLLM exception when using OpenAI Speech-To-Text models #8996
Description
What happened?
I'm trying to use DSPy for an audio transcription task. From what I see in the documentation, this should be supported with a dedicated dsp.yAudio input class.
When I execute the attached code example I receive the following error:
litellm.exceptions.BadRequestError: litellm.BadRequestError: OpenAIException - This is not a chat model and thus not supported in the v1/chat/completions endpoint. Did you mean to use v1/completions?
I do manage to use OpenAI's speech-to-text models using LiteLLM directly though:
from litellm import transcription
from pathlib import Path
audio_file = Path("example_001.m4a").open("rb")
response = transcription(model="gpt-4o-transcribe", file=audio_file)
print(f"response: {response}")Am I missing anything or is this a bug with the way LiteLLM is used behind the scenes?
Steps to reproduce
from pathlib import Path
import dspy
class TranscribeSignature(dspy.Signature):
audio: dspy.Audio = dspy.InputField()
transcription: str = dspy.OutputField()
if __name__ == "__main__":
audio_path = Path("example_001.m4a")
audio = dspy.Audio.from_file(audio_path)
lm = dspy.LM(model="gpt-4o-mini-transcribe")
dspy.configure(lm=lm)
transcriber = dspy.Predict(TranscribeSignature)
transcription = transcriber(audio=audio) # <- raises litellm.exceptions.BadRequestError
print(transcription)DSPy version
3.0.3