Open sxjscience opened 4 years ago
In GluonNLP, we have reached the consensus that the pretrained model files should be named as {NAME}-{8 character hash}.{ENDING}.
The reason is that when replacing a file, we should never break old code which relies on the old file.
We should later revise our contribution document to emphasize this.
I think this was the consensus and practice from the very inception of the project. Let's document it better
In GluonNLP, we have reached the consensus that the pretrained model files should be named as {NAME}-{8 character hash}.{ENDING}.
The reason is that when replacing a file, we should never break old code which relies on the old file.
We should later revise our contribution document to emphasize this.