The requirements contain flash-attn, which does not support Apple Silicon and Metal.

Running the examples without flash-attn runs into:

from transformers.models.whisper.modeling_whisper import WhisperEncoder, WhisperEncoderLayer, WhisperFlashAttention2
ImportError: cannot import name 'WhisperFlashAttention2' from 'transformers.models.whisper.modeling_whisper'

Is it planned? Any alternatives that can be easily integrated for ARM Mac users?

Might be helpful: whisper-mps.