Hi.
I have built a code submission similar to yours. However, my code does need to use PyTorch and Huggingface's transformers. But I am not sure how to do that in CodaLab environment. Could you give more info on how to modify the metadata file to install torch and other libs before running the program?
Thank you
Posted by: tuanlai245 @ Oct. 21, 2020, 3:19 p.m.Hi.
I have built a code submission similar to yours. However, my code does need to use PyTorch and Huggingface's transformers. But I am not sure how to do that in CodaLab environment. Could you give more info on how to modify the metadata file to install torch and other libs before running the program?
Thank you
Dear Tuan Manh Lai,
Thanks for your interest in our BioRelEx benchmark.
Before I provide specific metadata/environment files, can you indicate other requirements (hardware, in particular) for your submission?
Can your model run inference on CPU?
Hi Karen.
Thank you for your reply. My code does not have any specific hardware requirement. It can run on GPU.
So here are the module that my code requires:
+ PyTorch (1.4+)
+ transformers==3.0.2
+ pyhocon
+ boltons
Please provide my your submission ID, I'll try to show you what's wrong in your submission.
You have multiple submissions, I don't know which one to inspect.
Hi Karen. Could you help me check my submission #5?
Posted by: tuanlai245 @ Oct. 21, 2020, 7:03 p.m.First, in order to avoid very complex command in the "metadata" file,
you can point it to a separate "code.sh" which will
(1) prepare and pip install everything and then will
(2) run your codebase on the test set
Please have a look: https://git.io/JTuXS
Thank you very much for your help Karen. I will try your suggestion.
Posted by: tuanlai245 @ Oct. 21, 2020, 8:07 p.m.There's also an issue with the way you read the data.
Your code expects 'paperid' to be given, but expect only very necessary information is available for the samples from the test set.
BTW I couldn't find your model checkpoints/weights. I'll recommend that you submit the pretrained model, so your submission only needs to run inference on the test set.
Posted by: mahnerak @ Oct. 21, 2020, 8:12 p.m.Yes. Thank you for the info. I do have the model weight but it is quite heavy so I want to try make sure my base submission works first before uploading a version with the model weight.
Posted by: tuanlai245 @ Oct. 21, 2020, 8:15 p.m.Thank you. It works now. There is still some issue in my own code. But all the required libs have been installed successfully. Thank you Karen.
Posted by: tuanlai245 @ Oct. 21, 2020, 8:34 p.m.