The leftmost two columns in the table are the total tuning time of 2,000 trials each op and the final inference latency, respectively. With XGBoost tuner, I suppose the result after 2,000 trials is sufficient to illustrate the usability of selective tuning. Comparing to the full auto-tuning result could definitely be interesting, but it is time-consuming since the tuning space is the order of 10^5 for each op. I'll further study the tuning space coverage in the short future for sure, along with the dynamic shape codegen supports. (In fact, this RFC is more like a side application during the process of dynamic shape codegen investigation 😄 )
-- You are receiving this because you are subscribed to this thread. Reply to this email directly or view it on GitHub: https://github.com/dmlc/tvm/issues/4188#issuecomment-546220405
