slorado

Docker

Pre-built images are available on Docker Hub. The image bundles the slorado binary (on PATH, at /slorado/bin) along with the well tested models under /slorado/models. The image also bundles blue-crab and slow5tools.

Builds for both AMD (rocm) and NVIDIA (cuda) are available.

For AMD (rocm)

docker pull hasindu2008/slorado:0.6.0-rocm

Print the help message to check the image works:

docker run --rm hasindu2008/slorado:0.6.0-rocm slorado basecaller --help

To basecall on an AMD GPU, expose the GPU devices to the container and bind mount the directory holding your BLOW5 file:

docker run -v "$PWD":/data  \
  --device=/dev/kfd --device=/dev/dri \
  --security-opt seccomp=unconfined --group-add video \
  hasindu2008/slorado:0.6.0-rocm \
  slorado basecaller /slorado/models/dna_r10.4.1_e8.2_400bps_hac@v5.0.0 /data/reads.blow5 -o /data/reads.fastq -x cuda:all

Note that -x cuda:all selects all GPUs for the rocm build too. If --group-add video fails with Unable to find group video, pass the numeric gid instead ($(getent group video | cut -d: -f3)).

For NVIDIA (cuda)

docker pull hasindu2008/slorado:0.6.0-cuda

Print the help message to check the image works:

docker run --rm hasindu2008/slorado:0.6.0-cuda slorado basecaller --help

To basecall on an NVIDIA GPU, expose the GPU devices to the container and bind mount the directory holding your BLOW5 file:

docker run -v "$PWD":/data --gpus all \
  hasindu2008/slorado:0.6.0-cuda \
  slorado basecaller /slorado/models/dna_r10.4.1_e8.2_400bps_hac@v5.0.0 /data/reads.blow5 -o /data/reads.fastq -x cuda:all