Inference on Scala for Models trained on v1.0 is very slow


#1

We have been running inference using the Module API in Scala on mxnet v0.8. Recently we upgraded to v1.0 and the application has slowed down terribly (6 hrs -> 15hrs). Is there any modification that is expected on the client side when using the inference APIs?


#2

I believe it’s due to the removing of dispose() in GC: Fixing thread safety issues in Scala library

I’m trying to do a “real” fix.