Whisper-large-v3-tbv Endpoint

Beginners
1

Hello everyone,
I deployed an endpoint for whisper-large-v3-tbv. All statuses are running fine, but when I use the test playground or the default code, I always get a 404 error.
My endpoint looks like this:
https://xxxxx.us-east-1.aws.endpoints.huggingface.cloud
And the default parameters:

{
  "inputs": "%_BASE_64_AUDIO_DATA_%",
  "parameters": {}
}

I tried a mistral-7b endpoint and it works fine. But I can’t get the whisper one to work.
In the analytics panel, I can see the 404 errors.
I must be missing something. I looked all over the documentation but I don’t understand what’s wrong.
Can someone help me please?

1 Like
2

In the case of Endpoint for Whisper, it may be necessary to send raw audio instead of JSON like this.

# load audio file
wget /static-proxy?url=https%3A%2F%2Fcdn-media.huggingface.co%2Fspeech_samples%2Fsample1.flac

# run request
curl --request POST \
  --url https://{ENDPOINT}/ \
  --header 'Content-Type: audio/x-flac' \
  --header 'Authorization: Bearer {HF_TOKEN}' \
  --data-binary '@sample1.flac'

Or as a last resort, you can write handler.py to enable arbitrary actions…

3

Hi, thanks for your answer.
In my case, I’m using the test playground, It automatically replaces the input with the audio file data.
I still tried your example in command line, with wget and curl, but I still get a 404 error.

1 Like
4

Hmm, something’s strange. I understand if everything isn’t working, but only Whisper isn’t working…:thinking:
This is different from the error that appeared in ASR before…

You might want to inquire by email. Since it’s a dedicated endpoint, it will cost money to experiment…

5

I think I can help with the Whisper endpoint issue! It’s actually a common problem with audio models.

Do this quick fix.

Instead of..

json

{
  "inputs": "%_BASE_64_AUDIO_DATA_%",
  "parameters": {}
}

Try this one code

json

{
  "inputs": "your_base64_audio_here"
}

This happens 'coz Whisper models work a bit differently than text models like Mistral. They’re more somewhat choosy about the request format.

If that doesn’t work:

  1. Make sure your audio file is small (under 1 minute)
  2. Try with a simple WAV or MP3 file

That should solve the 404 error!

Hope it works. :blush:

1 Like
6

Hi, thanks for the suggestion, but unfortunately it’s the same :frowning:

I’m using the test playground with a 20-second file.
It looks like the endpoint is not working.

1 Like
7
  1. Try a smaller file first, maybe 5-10 seconds instead of 20.
  2. Check the audio format, WAV files usually work best.
  3. Wait a few minutes, sometimes Whisper endpoints take a bit longer to fully activate even when they show “running”.

If still not working: The endpoint might need to be redeployed. Whisper models sometimes have deployment issues that don’t show up in the status.

1 Like