I also got same problem. (In my case, n/a tasks are not tried altough I set higher trials).
However, the model passed the memory verificaiton error after I forced to tune n/a tasks. (With using previously tuned log).
But a new problem is occured. The latency of tunned resnet18 is almost same with pytorch version resnet18.
@Hzfengsy Could you tell me the reason (I already tried 10,000 trials)? As I know, tuned resnet18 should performs much better than pytorch version model as shown in the TVM papers.