e.g. tested with [`ghcr.io/premai-io/audio-to-text-whisper-tiny-cpu:1.0.1`](https://github.com/premAI-io/prem-services/pkgs/container/audio-to-text-whisper-tiny-cpu) ([source](https://github.com/premAI-io/prem-services/blob/main/a2t-whisper/docker/cpu/Dockerfile)): - `docker run`: takes ~60 sec and uses 25% of my CPU to process https://github.com/htrivedi99/prem-blogs/blob/main/quarterly-earnings-chatbot/nvidia_earnings_call.mp3 - `docker run --gpus all`: takes ~5sec and uses 100% of my GPU + 25% of my CPU... even though the image is labelled as `-cpu` (non-GPU) Clearly some python dependency is pulling in CUDA libs. ```sh $ docker run --rm -t ghcr.io/premai-io/audio-to-text-whisper-tiny-cpu pip list | grep nvidia nvidia-cublas-cu11 11.10.3.66 nvidia-cuda-cupti-cu11 11.7.101 nvidia-cuda-nvrtc-cu11 11.7.99 nvidia-cuda-runtime-cu11 11.7.99 nvidia-cudnn-cu11 8.5.0.96 nvidia-cufft-cu11 10.9.0.58 nvidia-curand-cu11 10.2.10.91 nvidia-cusolver-cu11 11.4.0.1 nvidia-cusparse-cu11 11.7.4.91 nvidia-nccl-cu11 2.14.3 nvidia-nvtx-cu11 11.7.91 ``` Need to remove this unnecessary bloat from CPU images.
e.g. tested with
ghcr.io/premai-io/audio-to-text-whisper-tiny-cpu:1.0.1(source):docker run: takes ~60 sec and uses 25% of my CPU to process https://github.com/htrivedi99/prem-blogs/blob/main/quarterly-earnings-chatbot/nvidia_earnings_call.mp3docker run --gpus all: takes ~5sec and uses 100% of my GPU + 25% of my CPU... even though the image is labelled as-cpu(non-GPU)Clearly some python dependency is pulling in CUDA libs.
$ docker run --rm -t ghcr.io/premai-io/audio-to-text-whisper-tiny-cpu pip list | grep nvidia nvidia-cublas-cu11 11.10.3.66 nvidia-cuda-cupti-cu11 11.7.101 nvidia-cuda-nvrtc-cu11 11.7.99 nvidia-cuda-runtime-cu11 11.7.99 nvidia-cudnn-cu11 8.5.0.96 nvidia-cufft-cu11 10.9.0.58 nvidia-curand-cu11 10.2.10.91 nvidia-cusolver-cu11 11.4.0.1 nvidia-cusparse-cu11 11.7.4.91 nvidia-nccl-cu11 2.14.3 nvidia-nvtx-cu11 11.7.91Need to remove this unnecessary bloat from CPU images.