EnricoBeltramo

I'm using tensorflow gpu 1.13 on Windows 10. Should be a problem? What is the reference version?

To install it under windows I did: - created a virtual env conda with python 3.7 with follow versions of torch: conda install pytorch==1.4.0 torchvision==0.5.0 cudatoolkit=10.1 -c pytorch - I...

add to command line option: --num_workers=0 similar issue: https://github.com/rusty1s/pytorch_geometric/issues/366

@Somedaywilldo did you find something? I need to do same thing.

Same issue. Isn't possible to use alternatives to Pinecone?

How can I avoid that build? There is a repo where I can get the whl prebuild for my configuration or know witch python version/CUDA/other library are expected? I'm working...

No, I had a previous version of accellerate. I updated to latest (0.16.0) and now all work correctly! thank you!

The issue looks disappear if I do follow changes: - added an health check on localbroker in order to delay start of other broker: healthcheck: test: ["CMD", "curl", "-f", "http://localhost:8888/health"]...

Removed follow line (that configure a user/password to connect mqtt, if I understood correctly): DOCKER_VERNEMQ_USER_INMIND: 'inmind'

Discovered that with previous changes the issue disappears for most time, but randomly appears again on start stop of dockers. There are a solution for that? Unfortunately, with a similar...