forked from pytorch/FBGEMM
-
Notifications
You must be signed in to change notification settings - Fork 0
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
scale_bias_last option to support TBE layout (pytorch#848)
Summary: Pull Request resolved: pytorch#848 This diff adds the followings to be used in quantized table batched embedding (TBE) * scale_bias_last : by default true which is the old fbgemm CPU embedding JIT'ed kernel behavior. If false, scale and bias appear at the beginning of each row and are in fp16 matching with TBE. If false, it can also take -1 indices (output from pruned embedding id mapping) * OutType can be fp16 * output_stride and input_stride support for int4/int2 embedding * Fix a bug related to masking for fp16 Reviewed By: jianyuh Differential Revision: D33430251 fbshipit-source-id: 59569f2b1ebf8cde40756fa3d7d013a61da6736d
- Loading branch information
1 parent
fa20cb3
commit 747fc4a
Showing
8 changed files
with
1,228 additions
and
934 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Oops, something went wrong.