bump: ⬆️ CUDA Version

[odulcy-mindee]

Fix #1743

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 96.55%. Comparing base (83f1bc5) to head (84325af).
Report is 2 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1769      +/-   ##
==========================================
- Coverage   96.56%   96.55%   -0.02%     
==========================================
  Files         164      165       +1     
  Lines        7895     7901       +6     
==========================================
+ Hits         7624     7629       +5     
- Misses        271      272       +1     

Flag	Coverage Δ
unittests	96.55% <ø> (-0.02%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[odulcy-mindee]

@felixdittrich92 I checked disk usage for both images, I don't think it's worthwhile to make a distinction for base image:

ubuntu                             22.04                                      97271d29cb79   2 months ago    77.9MB
nvidia/cuda                        12.4.0-base-ubuntu22.04                    adcf2ab568f0   7 months ago    244MB

So I only kept nvidia/cuda:12.2.0-base-ubuntu22.04 as base image.
Moreover, with the current docTR setup, torch always install nvidia packages, so I decided to do the same with TensorFlow, which leads to simplification in the CI.
I also set CUDA version to 12.2. It's an arbitrary choice. It should be compatible enough for most setup.

Tell me what you think about this setup, especially the suggested changes I made in pyproject.toml

[felixdittrich92]

@felixdittrich92 I checked disk usage for both images, I don't think it's worthwhile to make a distinction for base image:

ubuntu                             22.04                                      97271d29cb79   2 months ago    77.9MB
nvidia/cuda                        12.4.0-base-ubuntu22.04                    adcf2ab568f0   7 months ago    244MB

So I only kept nvidia/cuda:12.2.0-base-ubuntu22.04 as base image. Moreover, with the current docTR setup, torch always install nvidia packages, so I decided to do the same with TensorFlow, which leads to simplification in the CI. I also set CUDA version to 12.2. It's an arbitrary choice. It should be compatible enough for most setup.

Tell me what you think about this setup, especially the suggested changes I made in pyproject.toml

The pyproject.toml change is fine for me it restores the old behaviour before tf 2.18 and at the end people can still install the prefered tf version before installing doctr 👍

Yeah every CUDA 12.X should be fine :)

I'm also fine with the base image - (less CI jobs ^^) because it doesn't matter much (torch and tf are already really huge packages) - people can still build there own containers - or use OnnxTR if they want something more lightweight/hardware optimized

[felixdittrich92]

@odulcy-mindee

Thanks LGTM 👍

But one thing the mypy CI job fails now due to:

ERROR: Could not install packages due to an OSError: [Errno 28] No space left on device

😅

[felixdittrich92]

So the problem now with the updated dev dependencies is that it needs to much space for the Github standard runners:

Option 1: Increase the runners to some larger ones (ubuntu-latest-large for example everywhere in the CI where we need to install .[dev] - I think this would need a configuration if you don't use enterprise on the mindee github account)
Option 2: ?? 😅 (some hacky quick fixes would be possible - but personally i don't like one of them) 😅

[felixdittrich92]

Ok after cleaning the action caches it works again maybe we should add a CI job to auto clean the caches:

https://docs.github.com/de/actions/writing-workflows/choosing-what-your-workflow-does/caching-dependencies-to-speed-up-workflows#managing-caches

Failing mypy can be ignored the issues are already solved on main

[felixdittrich92]

For cache cleaning we could add a CI job (other PR)

For example:

name: Clear Cache

on:
  workflow_dispatch:

on:
  schedule:
    - cron: '0 0 * * *'  # Runs once a day
  
jobs:
  clear:
    name: Clear cache
    runs-on: ubuntu-latest
    steps:
    - uses: MyAlbum/purge-cache@v2
      with:
          max-age: 172800 # Caches older then 2 days are delete

[odulcy-mindee]

@felixdittrich92 Ok, seems a good idea to add a CI Job in another PR