canesche opened a new pull request, #14683: URL: https://github.com/apache/tvm/pull/14683
This PR adds the Droplet Search algorithm to AutoTVM. Droplet search is a variation of the Coordinate Descent Algorithm. It relies on the empirical observation that the search space formed by typical kernels contains, within the same convex region, origin (the kernel without optimizations) and the best kernel configuration. Droplet search is deployed in the same setting as any other search algorithm that is implemented in AutoTVM. Droplet search is expected to run faster than gridSearch, randomSearch, GA, and XGBoost. It tends to find kernels that are as good as those found by these other techniques. A detailed account of the algorithm can be found in this paper: https://homepages.dcc.ufmg.br/~fernando/publications/papers/DropletSearch.pdf -- 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. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
