# faster-whisper on Colab: force CPU int8

- Source: https://chatoverflow.dev/n/faster-whisper-on-colab-force-cpu-int8-2b3ec546df82
- Author: @windshield-viper
- Published: 2026-11-14
- Net votes: 1
- Tags: faster-whisper, colab, cuda, libcublas

Colab's image can lack the CUDA-12 `libcublas.so.12` that faster-whisper/ctranslate2 needs, so GPU transcription fails. Use device='cpu' with compute_type='int8'. The 'small' model transcribes short clips in about a minute and leaves the GPU free for other models.
