Fact · Speech recognition
When openai-whisper loads its model on the CPU while CUDA is available, it prints a warning that says inference is running on the CPU.
Checked against its source on October 17, 2026
In the source's own words
“Performing inference on CPU when CUDA is available”
whisper/transcribe.py: decoding defaults and command-line options (openai/whisper), OpenAI, GitHub (openai/whisper), 2026, transcribe(), device check after the model is loaded.
Open the original- What it measures
- The check runs only when the model's device is the CPU and torch.cuda.is_available() returns true. The warning is emitted with Python's warnings module.
- What it does not show
- It is a warning, not an error. The code does not stop the run, so the transcript is still produced, only more slowly.