Skip to content

Latest commit

 

History

History
27 lines (20 loc) · 813 Bytes

README_v2.md

File metadata and controls

27 lines (20 loc) · 813 Bytes
Raw

Quick start

Training

To train a model, run

python run_trainer.py \
  --data_path=/path/to/cifar10/binary/files \

Note that you can terminate training prematurely, and pick up where you left off by setting --ckpt_path= to the path to the directory containing all checkpoint files generated so far. The parameters will be restored from the most recent checkpoint. Also, the training metrics (loss and accuracy) will be written to ./log. Run tensorboard --logdir=log to view tensorboard.

Evaluation

To evaluate a model, run

python run_evaluator.py \
  --data_path=/path/to/cifar10/binary/files \
  --ckpt_path=/path/to/directory/ckpt/files/will/be/loaded/from

To see full list of arguments, run

python run_trainer.py --help
python run_evaluator.py --help