You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
in a sparsely gated mixture of experts where each expert has state (a memory), there is a KeyError if different experts are activated on apply as happened to be activated on init -- to solve this, you can pass an 'init' flag into your custom module and if it is True, then you just use all the experts on that call. If that's a memory issue, you can use them one by one. Just make sure init hits all conditional branches of the modules with state
(let me know if there's an easier solution)
The text was updated successfully, but these errors were encountered:
in a sparsely gated mixture of experts where each expert has state (a memory), there is a KeyError if different experts are activated on
apply
as happened to be activated oninit
-- to solve this, you can pass an 'init' flag into your custom module and if it is True, then you just use all the experts on that call. If that's a memory issue, you can use them one by one. Just make sureinit
hits all conditional branches of the modules with state(let me know if there's an easier solution)
The text was updated successfully, but these errors were encountered: