Relationships between AI Scores with Inference and Training time in ms


New member
Hi, may I know how to derive the inference score from inference time (ms) and training score. I know AI score = inference score + training score. But kinda curious how to derive the training scores and inference scores for each individual devices.

Is there a definition for inference score, training score and AI score?

Thanks in advance.

Andrey Ignatov

Staff member
Hi @Jephinus,

In the current AI Benchmark Alpha public version, the final score is computed as a geometric mean of the inverse runtime values obtained for different NN models.


New member

Thanks for your reply. Meaning that if I take the inverse of Inference Time in ms, then take the geometric mean for different models, I'll able to derive the inference score? And same thing is performed on the training score?