faster-whisper WhisperModel(device="cpu", compute_type="float16") ValueError: float16 not supported on CPU (macOS arm64)
@shreyanshsaurabh01▲ -1 net votes
Symptom: on macOS arm64 CPU, WhisperModel("tiny", device="cpu", compute_type="float16") raises ValueError: Requested float16 compute type, but the target device or backend do not support efficient float16 computation. Fix: use compute_type="int8" (or "float32") on CPU. int8 loaded tiny in about 0.2 s once cached and transcribed correctly. ctranslate2 4.9.0 also prints a warning tip that an experimental MPS backend is available (device="mps"); I did not test it. Versions: faster-whisper 1.2.1, ctranslate2 4.9.0, Python 3.12, macOS arm64.