How do I use the checkpoints from Google Cloud?

[BDHU]

This may sound trivial. But I'd like to have some idea on how use the .data and .index file from gs://vmoe_checkpoints. Does this repo provide any scripts or code to do some evaluation/inference on those models? Any help would be appreciated.

[jpuigcerver]

Hi, apologies for the late reply.

To run the fine-tuning scripts, you must simply pass the corresponding file to the main.py script:

python vmoe/main.py --workdir=/tmp/foo/bar --config=vmoe/configs/vmoe_paper/vmoe_s32_last2_ilsvrc2012_randaug_light1_ft_ilsvrc2012

[BDHU]

Thanks for the reply. I was able to get the fine-tuning script running. It seems to me the script is always doing training & eval. Is there any way to use the checkpoint in the workdir directly for inference? I had to hack the train_and_eval loop and disable backprop to get inference working without retraining. Any help would be appreciated:)

[jpuigcerver]

Unfortunately, we don't have a script for running inference alone. But it should be relatively easy to create one using the classes/functions in evaluator.py.

[Wolfboyccc]

May I ask if your evaluation function has been written and how is it written?