Using VESSL’s Python SDK, you can view and interact with metrics and media logged to TensorBoard directly on VESSL. We currently support scalars, images, and audio.

VESSL supports TensorBoard with TensorFlow, PyTorch (TensorBoard > 1.14), and TensorBoardX.

You can integrate TensorBoard by simply adding vessl.init(tensorboard=True)to your code.

Note that this should be called before creating the file writer. This is because VESSL auto-detects the TensorBoard logdir upon writer creation but cannot do so if the writer has already been created.

​
TensorFlow

import tensorflow as tf
import vessl

vessl.init(tensorboard=True) # Must be called before tf.summary.create_file_writer

w = tf.summary.create_file_writer("./logdir")
...

​
PyTorch

from torch.utils.tensorboard import SummaryWriter
import vessl

vessl.init(tensorboard=True) # Must be called before SummaryWriter

writer = SummaryWriter("newdir")
...

​
TensorBoardX

from tensorboardX import SummaryWriter
import vessl

vessl.init(tensorboard=True) # Must be called before SummaryWriter

writer = SummaryWriter("newdir")
...
KerasTransformers