Profiler : No Profile Data was Found in Collab code example
I am new to Tensorflow, and was trying to figure out how to use the profiler. By following the docs ( https://www.tensorflow.org/tensorboard/tensorboard_profiling_keras ), I got tensorboard working on my remote development machine and could view everything except the profiler output.
I then tried executing the code example mentioned in the doc in Collab ( https://colab.research.google.com/github/tensorflow/tensorboard/blob/master/docs/tensorboard_profiling_keras.ipynb ) only to see the same output.
Is the code example not up to date or is there an issue somewhere else?
Thanks

I am running into the same problem after running the command tensorboard --logdir logs/
System information:
- Azure ML compute instance
- Linux 22.08.19
Environment information
- python=3.8.16
- pip= 22.3.1
- tensorboard=2.10.1
- tensorboard-plugin-profile=2.10.1
Files in logs directory

Tensorboard:

I run into the same problem, too
I can confirm that I'm experiencing the same issue 🫤
same here. Does anybody know how to solve this? Several months have passed, without even a single reply from the developers.
Hey, just took a look, seems the log directory is missing an events file, we will look into fixing this. As a quick workaround for now, please try:
- download the events.out.tfevents file at https://github.com/tensorflow/profiler/tree/master/demo
- upload this file to the demo log directory like:
then you should be able to see the profiler:
When using the profiler, the logs address should include the run ID. In the original example, this would be: %tensorboard --logdir='logs/20230414-182044'
Hey, just took a look, seems the log directory is missing an events file, we will look into fixing this. As a quick workaround for now, please try:
- download the events.out.tfevents file at https://github.com/tensorflow/profiler/tree/master/demo
- upload this file to the demo log directory like:
then you should be able to see the profiler:
![]()
Hi, I have the same problem showing that "No profile data was found". However, there exists a file events.out.tfevents.1713334799.XX.profile-empty in folder logs/20240417-141935.
My TF version is also 2.10.1, but I try to profile everything in my own computer.
I even tried the advice from @lakeyk , but that still didn't work.
Seems that everything required by tensorboard is present:
│ events.out.tfevents.1713334799.XX.profile-empty
│
├─plugins
│ └─profile
│ └─2024_04_17_06_19_59
│ XX.input_pipeline.pb
│ XX.kernel_stats.pb
│ XX.memory_profile.json.gz
│ XX.overview_page.pb
│ XX.tensorflow_stats.pb
│ XX.trace.json.gz
│ XX.xplane.pb
│
├─train
│ events.out.tfevents.1713334775.XX.40004.0.v2
│
└─validation
events.out.tfevents.1713334787.XX.40004.1.v2
I'm not sure whether it is an issue of this version of TF or not though.
I recommend updating Tensorflow to the latest version. I am able to successfully load the files with the given file structure with version 2.16.1, but am not able to with version 2.10.1.
@lakeyk could you share a gist/colab notebook to reproduce it?
Note that the problem in most cases is:
- Not using the right
--logdirpath - Having correct logdir path but not showing Overview page
-
To solve number 1 check the correct
--logdiras adviced by @lakeyk, something like%tensorboard --logdir logs/IDin Colab, or w/o%sign from CLI. -
To solve number 2 for CPUs take a look at this code GIST, then Step-Time Graph should show up.
-
Note that for GPUs it won't show the Step-Time Graph 😭 but the Tools (dropdown menu) does show other stats. Also the framework ops are shown, but to be honest it's not clear what they show (probably Host to Device communication.)
then you should be able to see the profiler: