-
Notifications
You must be signed in to change notification settings - Fork 426
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add variable batch per feature support to EBC (tw/cw only) #1358
Conversation
This pull request was exported from Phabricator. Differential Revision: D48805440 |
This pull request was exported from Phabricator. Differential Revision: D48805440 |
) Summary: Pull Request resolved: pytorch#1358 X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and reverse_indices to the KJT Reindexing is done internally with EBC and with a user provided reverse indices tensor Differential Revision: D48805440 fbshipit-source-id: 3f9439b0ca5325489fcf9c629bffe95da53ece9c
9d3adb8
to
55da48d
Compare
) Summary: X-link: pytorch/torchrec#1358 Pull Request resolved: pytorch#1986 To enable provide stride_per_key_per_rank and reverse_indices to the KJT Reindexing is done internally with EBC and with a user provided reverse indices tensor Differential Revision: D48805440 fbshipit-source-id: 504f1dd0ebe0c2de28fe4d2bc623e3ba3b4da69a
This pull request was exported from Phabricator. Differential Revision: D48805440 |
) Summary: Pull Request resolved: pytorch#1358 X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and reverse_indices to the KJT Reindexing is done internally with EBC and with a user provided reverse indices tensor Differential Revision: D48805440 fbshipit-source-id: 2cb3665209025a44e659e5277534ad3f9e41ae0d
55da48d
to
b9d30a3
Compare
) Summary: X-link: pytorch/torchrec#1358 Pull Request resolved: pytorch#1986 To enable provide stride_per_key_per_rank and reverse_indices to the KJT Reindexing is done internally with EBC and with a user provided reverse indices tensor Differential Revision: D48805440 fbshipit-source-id: ded203e36f575e3ffb8a5b04591a1891bdccc1bc
This pull request was exported from Phabricator. Differential Revision: D48805440 |
) Summary: Pull Request resolved: pytorch#1358 X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and reverse_indices to the KJT Reindexing is done internally with EBC and with a user provided reverse indices tensor Differential Revision: D48805440 fbshipit-source-id: b151328e3d2e8ec6cae4f235db505d9a3a70e3fd
b9d30a3
to
8d5e8ec
Compare
) Summary: X-link: pytorch/torchrec#1358 Pull Request resolved: pytorch#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with a user provided inverse indices tensor we provide a utility function `merge_inverse_indices_per_key` to convert inverse indices in format `Dict[str, torch.Tensor]` to `torch.Tensor` accepted by KJT constructor Differential Revision: D48805440 fbshipit-source-id: cf889a87ab0f516897a7476268f321a97db079b5
This pull request was exported from Phabricator. Differential Revision: D48805440 |
) Summary: Pull Request resolved: pytorch#1358 X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with a user provided inverse indices tensor we provide a utility function `merge_inverse_indices_per_key` to convert inverse indices in format `Dict[str, torch.Tensor]` to `torch.Tensor` accepted by KJT constructor Differential Revision: D48805440 fbshipit-source-id: f66d4ec17457780922d8d63a8b71d45bd272c535
8d5e8ec
to
64bf686
Compare
) Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with a user provided inverse indices tensor we provide a utility function `merge_inverse_indices_per_key` to convert inverse indices in format `Dict[str, torch.Tensor]` to `torch.Tensor` accepted by KJT constructor Differential Revision: D48805440
64bf686
to
a712d9f
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
) Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with a user provided inverse indices tensor we provide a utility function `merge_inverse_indices_per_key` to convert inverse indices in format `Dict[str, torch.Tensor]` to `torch.Tensor` accepted by KJT constructor Differential Revision: D48805440
a712d9f
to
ebdfdd9
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
ebdfdd9
to
f00d38a
Compare
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Differential Revision: D48805440
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Differential Revision: D48805440
f00d38a
to
357a029
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
1 similar comment
This pull request was exported from Phabricator. Differential Revision: D48805440 |
This pull request was exported from Phabricator. Differential Revision: D48805440 |
Summary: X-link: pytorch/torchrec#1358 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
72d6c5d
to
9335ad6
Compare
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
This pull request was exported from Phabricator. Differential Revision: D48805440 |
Summary: X-link: pytorch/torchrec#1358 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
9335ad6
to
bdb1ae1
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
1 similar comment
This pull request was exported from Phabricator. Differential Revision: D48805440 |
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/torchrec#1358 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
bdb1ae1
to
c72e775
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
Summary: X-link: pytorch/torchrec#1358 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
c72e775
to
2fd175a
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
1 similar comment
This pull request was exported from Phabricator. Differential Revision: D48805440 |
Summary: X-link: pytorch/torchrec#1358 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/FBGEMM#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
2fd175a
to
766924f
Compare
This pull request was exported from Phabricator. Differential Revision: D48805440 |
Summary: X-link: pytorch/torchrec#1358 Pull Request resolved: pytorch#1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440 fbshipit-source-id: 9f49f7789ea2a20344cc56f84018fc761892836d
Summary: X-link: pytorch/torchrec#1358 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440
Summary: X-link: pytorch/torchrec#1358 Pull Request resolved: #1986 To enable provide stride_per_key_per_rank and inverse_indices to the KJT Reindexing is done internally with EBC and with user provided inverse indices Reviewed By: bigning Differential Revision: D48805440 fbshipit-source-id: e490baa41b9c91b02e6c246940d4a3a2c4551d65
Summary:
To enable provide stride_per_key_per_rank and reverse_indices to the KJT
Reindexing is done internally with EBC and with a user provided reverse indices tensor
Differential Revision: D48805440