README.md

TensorFlow Runtime Tracing Metadata Visualization

This library visualizes the runtime trace of TensorFlow.

Compare to built-in Chrome Tracing Format:

• Retains op execution details. (e.g., RecvTensor ops source and destination)
• Separates network ops from computing
• Faster Navigation

Example: timeline of Inception execution is available here

How to use

Given a collected runtime_metadata:

import tfvis
...
tfvis.timeline_visualize(runtime_metadata, "example.html")

How to Collect Runtime Metadata in TensorFlow

...

with tf.train.MonitoredTrainingSession(...) as sess:
	...
	run_metadata = tf.RunMetadata()
	options = tf.RunOptions(trace_level=tf.RunOptions.FULL_TRACE)
	sess.run(train_op, run_metadata=run_metadata, options=options)
...

Known Issues

• Has been tested only on Python 3
• Hover boxes may not be appropriately placed.

Cite this tool

@misc{Hashemi2018,
  author = {Sayed Hadi Hashemi},
  title = {TensorFlow Runtime Tracing Metadata Visualization},
  year = {2018},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/xldrx/tensorflow-runtime-metadata-visualization}},
}