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
set_allocator enables users to set the GPU allocator and configure RMM. It's a wrapper around rmm.reinitialize.
We should either add this to the documentation or deprecate it in favor of users directly using rmm.reinitialize. Searching for set_allocator in the docs returns no results.
The text was updated successfully, but these errors were encountered:
cudf.set_allocator is a wrapper around rmm.reinitialize and I think we should deprecate it in favor of the latter:
It brings more visibility to RMM, which is good to know about if you are using RAPIDS and care about memory management/performance.
In code that uses cuDF in combination with other RAPIDS libraries (e.g, cuml), using cudf.set_allocator() will implicitly configure memory management for other libraries as well, which feels smelly to me compared to using rmm.reinitialize() to configure memory management explicitly for all of RAPIDS.
set_allocator enables users to set the GPU allocator and configure RMM. It's a wrapper around
rmm.reinitialize
.We should either add this to the documentation or deprecate it in favor of users directly using
rmm.reinitialize
. Searching for set_allocator in the docs returns no results.The text was updated successfully, but these errors were encountered: