The armory-uploader places its files in eval7/CODE/addr/model_weights on S3. However, Armory expects them to be in a flat directory.
The notification will be for a path like s3://armory-submission-data/eval6/JHUM/pkan2@jhu.edu/model-weights/JHUM_deformable_detr_AT.pth where the eval number, the CODE, and the address will be variable.
Using this example the function will want to copy that file JHUM_deformable_detr_AT.pth to a common model weights directory. We have assumed that the filenames will be unique in the common directory, because they always have been. The copy destination folder is s3://armory-submission-data/model-weights/
Find a way such that the lamba triggers when the upload is complete, and then initiate the copy action. These files can be on order of 1GB so copy will likely not be quick.
You should be able to test the lambda function by copying files to s3://armory-submission-data/eval7/GARD/cdh@example.com/model-weights/` using the AWS CLI
The armory-uploader places its files in
eval7/CODE/addr/model_weightson S3. However, Armory expects them to be in a flat directory.The notification will be for a path like
s3://armory-submission-data/eval6/JHUM/pkan2@jhu.edu/model-weights/JHUM_deformable_detr_AT.pthwhere the eval number, the CODE, and the address will be variable.Using this example the function will want to copy that file
JHUM_deformable_detr_AT.pthto a common model weights directory. We have assumed that the filenames will be unique in the common directory, because they always have been. The copy destination folder iss3://armory-submission-data/model-weights/Find a way such that the lamba triggers when the upload is complete, and then initiate the copy action. These files can be on order of 1GB so copy will likely not be quick.
You should be able to test the lambda function by copying files to s3://armory-submission-data/eval7/GARD/cdh@example.com/model-weights/` using the AWS CLI