Fact · Speech recognition
When openai-whisper runs on a CPU, it warns that FP16 is not supported there and uses FP32 instead.
Checked against its source on October 17, 2026
In the source's own words
“FP16 is not supported on CPU; using FP32 instead”
whisper/transcribe.py: decoding defaults and command-line options (openai/whisper), OpenAI, GitHub (openai/whisper), 2026, Model decoding setup.
Open the original- What it measures
- FP16 is the default precision for the command line (--fp16 is True by default). The warning is the expected message on a CPU-only computer.
- What it does not show
- The warning is harmless, but it means the run uses about twice the memory of an fp16 run.