Open adam-peaston-SC opened 4 months ago
Addition of LAMB optimizer, CutMix & Mixup, BCE loss, and label smoothing.
ResNet50 x ImageNet
Global batch size of 2,160. Trained on 12 nodes. Learning rate 0.008. Trained for 300 epochs.
A gist of logs, a link to tensorboard etc if relevant.
General:
If adding a demo model training script:
Description
Addition of LAMB optimizer, CutMix & Mixup, BCE loss, and label smoothing.
Model details
ResNet50 x ImageNet
Steps to reproduce any training
Global batch size of 2,160. Trained on 12 nodes. Learning rate 0.008. Trained for 300 epochs.
Training results
A gist of logs, a link to tensorboard etc if relevant.
Things done
General:
If adding a demo model training script: