ERROR: Detected NVIDIA Tesla K80 GPU, which is not supported by this container, STT Docker Image about stt

[etlweather]

I have a different GPU (RTX 3070) but had the same issue. So I built a container by changing the Dockerfile.train base image (FROM) with nvcr.io/nvidia/tensorflow:21.12-tf1-py3. I had to make a few tweaks but it overall worked and I was able to create a new audio model using this container and my 4 GPUs.

But then to export it to TFLITE, I used the Docker image from the documentation as the latest TF1 does not support TFLITE.

[reuben]

The base image we use is an official NVIDIA image, so this is a bit surprising. You might want to try reverting the base image to tensorflow/tensorflow:1.15.4-gpu-py3 and seeing if that helps, but otherwise this is at most a documentation bug on our side.

[jmukiibi]

Alright, Thanks @reuben

[reuben]

@jmukiibi have you tested my suggestion? Did it work? Would be nice to know so we can include it in docs.

[jmukiibi]

@reuben I didn't try this out though. I ended up switching to a Tesla V100 GPU before I saw the response. Its worth a try. I will try to create the same instance and reproduce the error then try to see how it goes. Thanks