openai

/

whisper-small

like 153

please refer to the issue detailed here; https://discuss.huggingface.co/t/deploying-open-ais-whisper-on-sagemaker/24761/54?u=mllife

basically, the currently sagemaker have a max payload set to 5 MB, and there is no way around it on how the current code for whisper is streaming the file to the end-point using just audio_path as input; which makes this model nearly useless for sagemaker deployment.

if someone has done custom inference.py which loads file from s3_path at the endpoint itself and later processes it, please share