I have created a ML model using AI Center from 500 invoices and I have currently deployed а skill with a public endpoint for a DU process. The process has so far gone over around 1500 two page invoices. I use machine learning extractor and if something is missing- machine learning extractor trainer for those 2-5 invoices that don’t pass the auto extractor validation. I have also done around 10 full pipeline runs which should cost around 600 units because I ran them with GPU enabled.
The problem is that currently my 60k AI Unit are depleted. I don’t know how this could’ve occurred when I’ve spend around 3600 (1500x2 ai units + 600 for the pipeline runs) according to UiPath metrics.
Is there any explanation to why this might occur?
Also is there some way to check where all those units are wasted, like a log file?
Please reach out to your account manager or raise a support ticket to get more detailed information but one thing that may consume AI Units that is not in your computation above is hosting the ML Skill (assumng you are on Cloud).
If you are hosting on CPU (what we usually recommend), this by default cost 2 AI Units per hour, if you are hosting on GPU then it cost 40 AI Units per hour. So if the Skill is deployed on GPU it may increase the consumption pretty quickly.
See full documentation here: AI Units
Also make sure than no other project is also consuming AI Units.
I reached out to UiPath support and they told me the same thing. I had two skills both with GPU enabled which lead to the consumption of the units 20X faster than expected. I’ve disabled the ‘Enable GPU’ option now.