Can you share the pre-trained desktop GPU latency predictor and FLOPS predictor?

SamsungLabs / eagle

Measuring and predicting on-device metrics (latency, power, etc.) of machine learning models

Apache License 2.0

66 stars 12 forks source link

Can you share the pre-trained desktop GPU latency predictor and FLOPS predictor? #2

Closed auroua closed 3 years ago

auroua commented 3 years ago

To compare the performance of the binary relation predictor with our proposed predictor, can you share the pre-trained desktop GPU latency predictor and FLOPS predictor?

vaenyr commented 3 years ago

Hi, we've uploaded both predictors to the google drive. The latency predictor outputs normalised values so you will have to denormalise them using the following constants (they can be obtained from the provided dataset of desktop GPU latencies):

Max: 0.00951075005531311
Min: 0.0008091678619384765 We use a simple linear normalisation so you only need to scale by delta and add min. Feel free to reopen the issue if you have any other questions/problems. :)

auroua commented 3 years ago

@vaenyr Thanks!