Support INT4 Dequant onto GPU for Seq INT TBE look up #3584
Conversation
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
|
This pull request was exported from Phabricator. Differential Revision: D68187234 |
✅ Deploy Preview for pytorch-fbgemm-docs ready!
To edit notification comments on pull requests, go to your Netlify site configuration. |
faran928
added a commit
to faran928/FBGEMM
that referenced
this pull request
Jan 17, 2025
Summary: Seq INT4 -> INT4 STBE look up is supported in the diff stack: https://www.internalfb.com/diff/D61305978 . This diff supports: 1. The dequanitzation of INT4 -> INT4 STBE look up onto Cuda for all float types 2. Extends the dequantization of INT4 > INT4 STBE look up onto CPU for BF16 The main gap is to handle the dequant for the case when scale bias for INT4 quantized tensor is in the front. While for CPU, just need to add the dequantization for BF16 based on dtype. This will enable us to reduce the network overhead to remote embedding server as well as D2H data transfer from onto GPU host. Differential Revision: D68187234
faran928
force-pushed
the
export-D68187234
branch
from
004b36b to
867b7f7
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68187234 |
Summary: Seq INT4 -> INT4 STBE look up is supported in the diff stack: https://www.internalfb.com/diff/D61305978 . This diff supports: 1. The dequanitzation of INT4 -> INT4 STBE look up onto Cuda for all float types 2. Extends the dequantization of INT4 > INT4 STBE look up onto CPU for BF16 The main gap is to handle the dequant for the case when scale bias for INT4 quantized tensor is in the front. While for CPU, just need to add the dequantization for BF16 based on dtype. This will enable us to reduce the network overhead to remote embedding server as well as D2H data transfer from onto GPU host. Differential Revision: D68187234
faran928
force-pushed
the
export-D68187234
branch
from
867b7f7 to
cd85b52
Compare
|
This pull request was exported from Phabricator. Differential Revision: D68187234 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
Seq INT4 -> INT4 STBE look up is supported in the diff stack: https://www.internalfb.com/diff/D61305978 .
This diff supports:
The main gap is to handle the dequant for the case when scale bias for INT4 quantized tensor is in the front. While for CPU, just need to add the dequantization for BF16 based on dtype.
This will enable us to reduce the network overhead to remote embedding server as well as D2H data transfer from onto GPU host.
Differential Revision: D68187234