Do you know if these implementations also support leveraging the M1/M2 GPU, such as shown here? https://github.com/openai/whisper/pull/382