How much VRAM is needed for finetuning the large model

[dgoryeo]

Would finetuning Whisper require significantlt more VRAM?
I am planning for finetuning the large-v2 model. I have two use case scenraios:

• (a) Finetuning for additional vocabolary of say 200,000 words.
• (b) Finetuning for additional language dataset

Any pointers or rough range would be helpful.

[glangford]

fyi, see

Memory requirements for local training
https://huggingface.co/openai/whisper-large-v2/discussions/21

[dgoryeo]

Thanks @glangford !!!