Dataset row limit for Benchmark – NanoEdge AI Studio official specification inquiry

[liu_pzh]

I would like to clarify the data constraints for Benchmark workflow in NanoEdge AI Studio:

What is the maximum supported number of lines for each imported CSV dataset during Benchmark?

Is there any official documented hard limit on the number of rows, or only practical recommendations?

Some community references mention a rough threshold of ~100,000 lines, and files exceeding this size may cause memory overflow, import truncation or failed Benchmark launch. Could you confirm whether this threshold is formally defined?

[Julian E.]

Hi ​@liu_pzh,

 

There is a limit for the size of raws, which is around 16000 if I remember correctly.

Honestly, I have never seen any project requiring this much input size.

 

Regarding the size of the dataset, the only limit is your RAM.

As we are doing cross validation, the dataset will be copied so the real size required is bigger than the size of your dataset.

 

In the last release, we implemented a way to select only a part of your dataset to avoid issues with user with datasets too big. It also improves the results of the in a lot of cases as we extract more relevant data instead of sending everything.

The models in NanoEdge are quite small, so the information learnt is limited and more data is not always equal to better results.

 

Please find some information here:

How exactly are datasets larger than 250 MB reduced during benchmarking? What sampling strategy is used? | Community

 

Have a good day,

Julian

[Carlo Ramirez]

As far as I know, there isn't an officially documented maximum row limit. It seems to be more of a practical limitation based on memory usage. If you're hitting issues, testing with smaller datasets is probably the best workaround until ST provides an official answer.

[Julian E.]

Hi ​@Carlo Ramirez,

 

Yes exactly.

Depending on your PC, a dataset too big could require more RAM / disk size than what you have or trigger a timeout.

To help with this, in the newest version, we now sort your dataset as explain in the link above, to reduce your dataset size and hopefully gives more relevant data to learn in order to improve the results.

With the Validation step, you can still test any dataset (complete ones).

 

Have a good day,

Julian

[Carlo Ramirez]

Thanks, Julian. That makes sense. I’ll keep the dataset size and system resources in mind. Appreciate the clarification!