torchrec
torchrec copied to clipboard
Pytorch domain library for recommendation systems
Summary: X-link: https://github.com/pytorch/FBGEMM/pull/3857 X-link: https://github.com/facebookresearch/FBGEMM/pull/949 Original commit changeset: e175eecac4c5 Original Phabricator Diff: D71444136 Reviewed By: sryap Differential Revision: D71578251
Summary: This diff reverts D74293458 T224582561 broke tests Reviewed By: PoojaAg18 Differential Revision: D74902138
Summary: When running torch.jit.script on the various forward functions, you can run into issues if there are any other utilites interacting with the function definitions. As an example, if you...
Summary: Update the `_maybe_compute_stride_kjt` logic to calculate stride based off of `inverse_indices` for VBE KJTs. Currently, stride of VBE KJT with `stride_per_key_per_rank` is calculated as the max "stride per key"....
Summary: Once, we TW shard the SSD tables here: D74444699 along with remaining tables, we traverse the model def and swap cpu tbe look up kernel for these ssd tables...
We are using DLRM model for personalization and we are getting CUDA error. By setting up CUDA_LAUNCH_BLOCKING flag and enabling cuda core dump, it pointed to two files where the...
Summary: # Context * Currently torchrec IR serializer does not support exporting variable batch KJT, because the `stride_per_rank_per_rank` and `inverse_indices` fields are needed for deserializing VBE KJTs but they are...