Skip to content

Latest commit

 

History

History
33 lines (29 loc) · 905 Bytes

docker-image.md

File metadata and controls

33 lines (29 loc) · 905 Bytes
Raw

Docker image

A Docker image with an installed perceiver-io library is available on the GitHub Container registry. Training runs can be started with:

sudo docker run \
  -v $(pwd)/.cache:/app/.cache \
  -v $(pwd)/logs:/app/logs \
  --rm \
  --ipc=host \
  --name=perceiver-io \
  --runtime=nvidia \
  ghcr.io/krasserm/perceiver-io:latest \
  python -m SCRIPT fit [OPTIONS]

where SCRIPT must be replaced by the module name of a training script and [OPTIONS] with the training script options. For example:

sudo docker run \
  -v $(pwd)/.cache:/app/.cache \
  -v $(pwd)/logs:/app/logs \
  --rm \
  --ipc=host \
  --name=perceiver-io \
  --runtime=nvidia \
  ghcr.io/krasserm/perceiver-io:latest \
  python -m perceiver.scripts.text.mlm fit \
    --model.params=deepmind/language-perceiver \
    ...