Evgenii Zheltonozhskii
Evgenii Zheltonozhskii
But batch size for 3090 is actually __lower__ than for M1? You could put numbers for best performance in both cases (i.e. run optimal bs for M1 on both and...
Cool! I think having best-performing results for both platforms would give people understanding what can you achieve in terms of peak performance.
If you run the code you'll see tqdm progress bar which shows average time per batch, elapsed time and approximate time to finish. Second progress bar shows same for epochs....
Why loss should be between 0-1?
cross entropy is not bounded, thus no, you're not right
You're very welcome to send a PR with adding the networks to examples
@shyhuai You should ask @tfboyd for optimal code.
@shelhamer @KeDengMS @piiswrong @soumith Sorry if you're wrong people to tag. Do you have something to add? Do think the benchmark can be improved somehow or your framework isn't used...
https://github.com/Randl/tdesktop/commit/7554b45080e93e5860ee4618ce9dfc2eac298ab7 List of headers: - qcoreapplication_p.h - qcssparser_p.h - qfixed_p.h - qfontengine_p.h - qfont_p.h - qfragmentmap_p.h - qglobal_p.h - qguiapplication_p.h - qharfbuzz_p.h - qobject_p.h - qplatformnativeinterface.h - qshortcutmap_p.h - qtextdocument_p.h...