Issue when uploading the models

Hi,

I encountered some issues while uploading my models. After compressing the folder containing the model weights and configuration files into a tar.gz format for the upload, my inference code can still not find the folder under /opt/ml/model/. I would like to ask if there are any file type restrictions or if there could be naming issues when decompressing.

Thank you.

Hi,

There is no special restrictions. My recommendation is to only convert the model weights to ".tar.gz". you can use the following code I used. and keep your other codes and config files in work directory or source folder.

import tarfile
import os

# Path where model saved
model_pth_path = os.path.join(cfg.OUTPUT_DIR, "model_final.pth")

# .tar.gz output path
tar_gz_path = os.path.join(cfg.OUTPUT_DIR, "model_final.tar.gz")

# Compress the .pth file into a tar.gz archive
with tarfile.open(tar_gz_path, "w:gz") as tar:
    tar.add(model_pth_path, arcname=os.path.basename(model_pth_path))

Grand-Chalenge will unzip your model into "/opt/ml/model/model_final.pth" automatically and you can reach it (doesn't need to unzip it)

please let me know if it works