Show HN: A Layer for robotics dataset quality

calibrarobotics.com

5 pointsomertt275 comments

We are exploring how teams can automatically audit data quality, detect problematic demonstrations, and build smaller, high-quality training sets before spending GPU time.

We are looking for design partners to help shape the Calibra.

Open articleView on HN

It's really impressive to see the quality of the datasets again; in most AIs, a bad dataset means that nothing works.

But how do you make it better for the GPU if the size is too large?

Thanks for the answer.We are approaching this with reducing unnecessary data rather than making the GPU handle more.So it analyzes quality, coverage, and redundancy to identify a smaller subset of demonstrations to use less data and compute costs.

How do you measure them ?

you can check our repo : https://github.com/omertt27/Calibra

Reply on news.ycombinator.com