[Bug]: groq/whisper-large-v3 returns 400 BadRequestError with OPENAI_TRANSCRIPTION_PARAMS

[afreenss]

What happened?

I'm using the groq/whisper-large-v3 model with litellm.transcription() as shown in the official documentation

This same code was running fine on Google Colab around 3–4 days ago, but now it consistently fails with a BadRequestError. No changes were made to the code or environment.

import os
from litellm import transcription

os.environ["GROQ_API_KEY"] = "your-key-here"

audio_file = open("/path/to/audio.mp3", "rb")

transcript = transcription(
    model="groq/whisper-large-v3",
    file=audio_file,
    prompt="Specify context or spelling",
    temperature=0,
    response_format="json"
)

print("response=", transcript)

Error message:

BadRequestError: Error code: 400 - {'error': {'message': 'unknown paramOPENAI_TRANSCRIPTION_PARAMS[]', 'type': 'invalid_request_error'}}

Relevant log output

9 audio_file = open(audio_path, "rb")
---> 11 transcript = litellm.transcription(
     12     model="groq/whisper-large-v3",
     13     file=audio_file,
     14     prompt="Specify context or spelling",
     15     temperature=0,
     16     response_format="json"
     17 )
     19 transcribed_text = transcript.text

File d:\agenticvenv\Lib\site-packages\litellm\utils.py:1283, in client.<locals>.wrapper(*args, **kwargs)
   1279 if logging_obj:
   1280     logging_obj.failure_handler(
   1281         e, traceback_exception, start_time, end_time
   1282     )  # DO NOT MAKE THREADED - router retry fallback relies on this!
-> 1283 raise e

File d:\agenticvenv\Lib\site-packages\litellm\utils.py:1161, in client.<locals>.wrapper(*args, **kwargs)
   1159         print_verbose(f"Error while checking max token limit: {str(e)}")
   1160 # MODEL CALL
...
-> 1037         raise self._make_status_error_from_response(err.response) from None
   1039     break
   1041 assert response is not None, "could not resolve response (should never happen)"

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

litellm==1.72.0

Twitter / LinkedIn details

No response

[jerry-intrii]

Same problem here. Downgrade to 1.71.1 works fine

[afreenss]

Yes. Downgrading to litellm 1.71.1 works fine for me as well 👍

[lishtys]

still not working in latest version

[jerry-intrii]

`File "/usr/lib/python3.13/site-packages/litellm/utils.py", line 1356, in wrapper_async

result = await original_function(*args, **kwargs)

         ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

File "/usr/lib/python3.13/site-packages/litellm/main.py", line 4756, in atranscription

raise exception_type(

      ~~~~~~~~~~~~~~^

    model=model,

    ^^^^^^^^^^^^

...<3 lines>...

    extra_kwargs=kwargs,

    ^^^^^^^^^^^^^^^^^^^^

)

^

File "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/exception_mapping_utils.py", line 2271, in exception_type

raise e

File "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/exception_mapping_utils.py", line 369, in exception_type

raise BadRequestError(

...<6 lines>...

)

litellm.exceptions.BadRequestError: litellm.BadRequestError: GroqException - unknown param GROQ_TRANSCRIPTION_PARAMS[]. Received Model Group=whisper-large-v3`

This is what I found in my log related to this issue's error report

[nkleemann]

same problem here

[jerry-intrii]

Error message:

500: litellm.BadRequestError: GroqException - unknown param atranscription. Received Model Group=whisper-large-v3
Available Model Group Fallbacks=None LiteLLM Retried: 1 times, LiteLLM Max Retries: 2

API Request:

curl -X POST \
  https://api.groq.com/openai/v1/ \
  -H 'Content-Type: application/json' \
  -d '{
  {
    "model": "whisper-large-v3",
    "file": {},
    "extra_body": {
      "model": "whisper-large-v3",
      "prompt": null,
      "temperature": null,
      "OPENAI_TRANSCRIPTION_PARAMS": [
        "language",
        "response_format",
        "timestamp_granularities"
      ],
      "atranscription": true
    }
  }
  }'

Need to remove the atranscription parameter, but not very familiar with other uses of atranscription.

[jerry-intrii]

@ishaan-jaff It would be kind if you could provide some instructions here.

[MattCarneiro]

any news?

[BrianToro]

Hi! the same problem here, would be good to hear any news @ishaan-jaff

[kanzure]

Instead of using the openai library to handle the HTTP request, why not use the groq library? https://console.groq.com/docs/speech-to-text

import os
import json
from groq import Groq

# Initialize the Groq client
client = Groq()

# Specify the path to the audio file
filename = os.path.dirname(__file__) + "/YOUR_AUDIO.wav" # Replace with your audio file!

# Open the audio file
with open(filename, "rb") as file:
    # Create a transcription of the audio file
    transcription = client.audio.transcriptions.create(
      file=file, # Required audio file
      model="whisper-large-v3-turbo", # Required model to use for transcription
      prompt="Specify context or spelling",  # Optional
      response_format="verbose_json",  # Optional
      timestamp_granularities = ["word", "segment"], # Optional (must set response_format to "json" to use and can specify "word", "segment" (default), or both)
      language="en",  # Optional
      temperature=0.0  # Optional
    )

    # To print only the transcription text, you'd use print(transcription.text) (here we're printing the entire transcription object to access timestamps)
    print(json.dumps(transcription, indent=2, default=str))

[MattCarneiro]

Instead of using the openai library to handle the HTTP request, why not use the groq library? https://console.groq.com/docs/speech-to-text

import os
import json
from groq import Groq

# Initialize the Groq client
client = Groq()

# Specify the path to the audio file
filename = os.path.dirname(__file__) + "/YOUR_AUDIO.wav" # Replace with your audio file!

# Open the audio file
with open(filename, "rb") as file:
    # Create a transcription of the audio file
    transcription = client.audio.transcriptions.create(
      file=file, # Required audio file
      model="whisper-large-v3-turbo", # Required model to use for transcription
      prompt="Specify context or spelling",  # Optional
      response_format="verbose_json",  # Optional
      timestamp_granularities = ["word", "segment"], # Optional (must set response_format to "json" to use and can specify "word", "segment" (default), or both)
      language="en",  # Optional
      temperature=0.0  # Optional
    )

    # To print only the transcription text, you'd use print(transcription.text) (here we're printing the entire transcription object to access timestamps)
    print(json.dumps(transcription, indent=2, default=str))

I am currently using groq and got the same error. Any news?

[bgeneto]

This error occurs when using GroqCloud with both openai-compatible provider (OpenAIException - unknown param OPENAI_TRANSCRIPTION_PARAMS[]) as in:

  - model_name: whisper-large-v3-turbo
    litellm_params:
      model: openai/whisper-large-v3-turbo
      api_key: os.environ/GROQ_API_KEY
      api_base: 'https://api.groq.com/openai/v1'
      additional_drop_params: ["GROQ_TRANSCRIPTION_PARAMS", "OPENAI_TRANSCRIPTION_PARAMS"]
      allowed_openai_params: ["model", "response_format"]

or directly via groq provider (GroqException - unknown param GROQ_TRANSCRIPTION_PARAMS[]):

  - model_name: whisper-large-v3-turbo
    litellm_params:
      model: groq/whisper-large-v3-turbo
      api_key: os.environ/GROQ_API_KEY

The worst part is that neither allowed_openai_params nor additional_drop_params actually works as a workaround. I've got GroqException - unknown param additional_drop_params[], it seems to be a different a bug in using additional_drop_params with litellm proxy.

So until someone merges a PR, transcription with Groq is pretty much dead in the water.