Persistent HTTP 500 on /v1/audio/transcriptions - "Unknown error when running prediction service

SambaNova Devs
1

Hi SambaNova Team/Community,

I am encountering a consistent Internal Server Error (500) when using the transcription endpoint. This setup was working perfectly about two weeks ago, but currently, every attempt fails.

The Details:

  • Model: Whisper-Large-v3

  • File Size: Tested with small files (3.5MB and smaller).

  • Auth: /v1/models returns 200 OK, so the API key is valid.

  • Implementation: Using the OpenAI Python SDK (v1.x.x) with base_url="``https://api.sambanova.ai/v1``".

  • Troubleshooting done: I have tried wrapping the audio in io.BytesIO with an explicit .name attribute and also tried a raw requests POST call to rule out SDK issues. Both methods result in the same 500 error.

The error message specifically mentions: "Something went wrong! Unknown error when running prediction service."

This suggests a failure at the RDU/Inference worker level rather than a request formatting issue. Are there known stability issues with the Whisper workers currently, or has the required schema for audio/transcriptions changed recently?

Any help or status update would be appreciated!

Log info:
2026-01-04 10:45:25,040 - INFO - HTTP Request: GET https://api.sambanova.ai/v1/models “HTTP/1.1 200 OK”
2026-01-04 10:45:25,095 - INFO - OpenAI client initialized successfully.
2026-01-04 10:45:32,341 - INFO - User initiated transcription for file: audio_sample.mp3
2026-01-04 10:45:32,343 - INFO - Sending transcription request for audio_sample.mp3…
2026-01-04 10:45:46,001 - INFO - HTTP Request: POST https://api.sambanova.ai/v1/audio/transcriptions “HTTP/1.1 500 INTERNAL SERVER ERROR”
2026-01-04 10:45:46,003 - INFO - Retrying request…
2026-01-04 10:46:00,132 - INFO - HTTP Request: POST https://api.sambanova.ai/v1/audio/transcriptions “HTTP/1.1 500 INTERNAL SERVER ERROR”
2026-01-04 10:46:14,469 - ERROR - SambaNova API Error: Something went wrong! Unknown error when running prediction service.

Traceback (most recent call last):
File “transcriber.py”, line 39, in run_transcription
response = self.client.audio.transcriptions.create(**params)
…
openai.InternalServerError: Something went wrong! Unknown error when running prediction service.

2

Hi @slicksroboto ,

Thank you for the details. We’ll work on replicating the issue and will get back to you with a solution as soon as possible.

Thanks,
Tanish