May be people have already figured it , so I like to know
Whether internet is enable so the code can download pre train weights.
My first submission says it is disabled.
If it is disabled then can we include weights in out gitlab repo, whether they will be copied.
Or pretrain models other than torchvision is not allowed, what are the rules.
If it is enable then I like to share my logs to organizers, how I can do that. Logs are saying that there is connection timeout
In this challenge, locally-trained models are not allowed.
Your code will be doing the training phase as part of AIcrowd submission as well.
Now facing cuda is not available
/srv/conda/envs/notebook/lib/python3.7/site-packages/torch/cuda/amp/grad_scaler.py:115: UserWarning: torch.cuda.amp.GradScaler is enabled, but CUDA is not available. Disabling.
warnings.warn(“torch.cuda.amp.GradScaler is enabled, but CUDA is not available. Disabling.”)
tried to submit 1 public code. Now this time got following
Time left: 10800
timeout: the monitored command dumped core
/home/aicrowd/run.sh: line 62: 365 Segmentation fault timeout $AICROWD_TIMEOUT_INFO python aicrowd_client/launcher.py
now submission are over.
It seems need to spend time on how to submit and not on the problem.
You can debug these issues locally by building a docker container. But you’re right, having to do this is not nice. I myself spent also more time than I would have liked on these issues.
Every code has its own package and dependency requirements. To simplify the process for our participants, we use forked version of
repo2docker. As the name suggests repo2docker fetches a git repository and builds a container image based on the configuration files found in the repository.
The configuration files need to be present in your repository root. The following configuration files can be used:
environment.yml - Install a conda environment (recommended)
Pipfile and/or Pipfil…