uvarc/tensorflow repository overview

⁠TensorFlow

https://tensorflow.org/⁠

⁠Preinstalled Python Packages

Bold: contained in all tags

To install more packages:

./<sif> -m pip install --user <package>

⁠Special note for 2.3.0 & 2.4.0 - "Distroless"

We experimented with the distroless base image in versions 2.3.0 and 2.4.0, which requires singularity 3.6+. Go through the next section to decide which tag (x.y.z, x.y.z-cpu, x.y.z-distroless) to pull.

We recommend x.y.z for use on Rivanna.

Since there is no shell in Distroless images, you will not be able to execute, say, ! in a Jupyter notebook.

⁠Which tag should I use?

Move right if "no"; down if "yes"

• Do you intend to use this on a GPU? --(no)--> x.y.z-cpu
• Do you need preinstalled Python packages? --(no)--> x.y.z-distroless
• x.y.z

⁠Details

• x.y.z-cpu
  • can only run on CPU
  • all tags without cpu can run on GPU
  • usage on Rivanna:

    module load singularity
    singularity pull docker://uvarc/tensorflow:x.y.z-cpu
    ./tensorflow_x.y.z-cpu.sif your_script.py
    

    Copy
• x.y.z-distroless
  • contains necessary CUDA libraries
  • usage on Rivanna:

    module load singularity
    singularity pull docker://uvarc/tensorflow:x.y.z-distroless
    singularity run --nv tensorflow_x.y.z-distroless.sif your_script.py
    

    Copy
• x.y.z
  • extends x.y.z-distroless with preinstalled Python packages (see below)
  • usage on Rivanna:

    module load singularity
    singularity pull docker://uvarc/tensorflow:x.y.z
    singularity run --nv tensorflow_x.y.z.sif your_script.py
    

    Copy