Deep Learning Profiler 19.10 Release Notes
Deep Learning Profiler 19.10 Release Notes (PDF)
Deep Learning Profiler (DLProf) is a tool for profiling deep learning models to help data scientists understand and improve performance of their models visually via Tensorboard or by analyzing text reports. It also helps understand resource usage when models are trained.
The key features of DLProf v0.5.0/r19.10:
- GraphDef file generation
- The GraphDef file is now automatically generated whenever you invoke DLProf.
- You can still specify a pre-generated GraphDef file.
- Full XLA Support
- You can now profile XLA compiled models to get maximum performance and see the profile output.
- Reports and graphs will partially map back to the original, pre-XLA, graph.
- Tensorboard Improvements:
- The Graph panel displays the correlation of an XLA cluster map to the original TensorFlow graph.
- Reports now include Tensor shapes, dimensions, and data types.
- This software is only accessible in the NGC TensorFlow container.
- This software is only supported by TensorFlow 1.14