Observer007 opened a new issue #5610: URL: https://github.com/apache/incubator-tvm/issues/5610
I tune a 1x1 conv with shape 28*128*56*56 and got a time cost gap between autotvm tuning log and actual running. The gap is about 0.01-0.05 ms and total execute time is 0.7-0.8 ms. I used autoTVM tuning and got up to 3800 gflops, but during running with best parameters, I just got 3600 gflops. The target and other parameters are same during tuning and running. My machine is intel clx-8280. I used time_evaluation api and read its source code. I find that the call of RPC and other target are different. AutoTVM uses rpc and my running uses llvm. So is this the case? ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
