search

NVIDIA-Merlin / HugeCTR

HugeCTR is a high efficiency GPU framework designed for Click-Through-Rate (CTR) estimating training
Apache License 2.0
937 stars 200 forks source link

[Requirement] Custom allocator support for gpu_cache #454

Open mfbalin opened 3 months ago

mfbalin commented 3 months ago

Is your feature request related to a problem? Please describe. We utilize gpu_cache in the Deep Graph Library project. We would like the gpu_cache to be able to use a custom allocator. That way, we can make it use the pytorch allocator in our project.

yingcanw commented 3 months ago

Thanks for your feedback. We have abstracted the allocator in the latest EmbedCache. You can implement a custom allocator for host and device memory allocation.