search

jina-ai / executors

internal-only
Apache License 2.0
31 stars 12 forks source link

feat: add gpu support to executors #128

Closed jakobkruse1 closed 3 years ago

jakobkruse1 commented 3 years ago

This PR adds GPU support and tests to compatible executors.

cristianmtr commented 3 years ago

Make sure to also test the docker image as I am doing in docker-tests

jakobkruse1 commented 3 years ago

Make sure to also test the docker image as I am doing in docker-tests

Do you mean just starting it with the default entry point?

tadejsv commented 3 years ago

Not to be merged before #125

tadejsv commented 3 years ago

One more thing to do here: now that we have GPU and CPU images, let's create separate requirements - should enable our CPU images to be substantially smaller.

For pytorch, I think you can do this by having requirements.txt like this (check that this is true)

-f https://download.pytorch.org/whl/torch_stable.html
torch==1.9.0+cpu

I would test this locally first, of course, and then I would add some step in CI that tells you the size of the docker image

tadejsv commented 3 years ago

Let's stop at the executors we have now - adding more can be done in other PRs. The main goal of this one is to get the infrastructure for GPU CI/CD in plance, the rest can follow

nan-wang commented 3 years ago

Let's stop at the executors we have now - adding more can be done in other PRs. The main goal of this one is to get the infrastructure for GPU CI/CD in plance, the rest can follow

I am totally agreed on this.

nan-wang commented 3 years ago

As for the failing tests on paddlepaddle, I think the main issue is the missing dependency of opencv-python

nan-wang commented 3 years ago

Another issue is about black, https://github.com/jina-ai/executors/pull/128/checks?check_run_id=3464128680#step:7:31

jakobkruse1 commented 3 years ago

Ready to merge - Reviews would be nice

jakobkruse1 commented 3 years ago

Party hard