Skip to content

Comments

[NPU] Add HcclBackend for 1-bit adam, 1-bit lamb, 0/1 adam#4733

Merged
conglongli merged 10 commits intodeepspeedai:masterfrom
RUAN-ZX:master-1bit-hcclbackend
Dec 9, 2023
Merged

[NPU] Add HcclBackend for 1-bit adam, 1-bit lamb, 0/1 adam#4733
conglongli merged 10 commits intodeepspeedai:masterfrom
RUAN-ZX:master-1bit-hcclbackend

Conversation

@RUAN-ZX
Copy link
Contributor

@RUAN-ZX RUAN-ZX commented Nov 27, 2023

To support NPU devices fulfilling feature requirements like 1-bit Adam, 1-bit Lamb, 0/1 Adam, I add HcclBackend and its corresponding import logics.
See what we have already done in #4567 .

@RUAN-ZX
Copy link
Contributor Author

RUAN-ZX commented Nov 27, 2023

@conglongli @awan-10 Would you please take a look? Thank you :)

@RUAN-ZX
Copy link
Contributor Author

RUAN-ZX commented Dec 1, 2023

@tjruwase, hi, could you review this pr?

@RUAN-ZX
Copy link
Contributor Author

RUAN-ZX commented Dec 7, 2023

@conglongli @awan-10 @tjruwase Could you please review this PR, cause we have some following PRs based on it. If I have done something wrong, please tell me, I will be happy to solve it. Thank you :)

@conglongli conglongli self-assigned this Dec 8, 2023
@conglongli
Copy link
Contributor

@RUAN-ZX Since we don't have the hardware, we will only review/approve based on whether your PR pass all our unit tests (it already failed the formatting test). And we would have to redirect all related future bug reports to you.

@CurryRice233
Copy link
Contributor

@RUAN-ZX Since we don't have the hardware, we will only review/approve based on whether your PR pass all our unit tests (it already failed the formatting test). And we would have to redirect all related future bug reports to you.

Of course, we'll fix all the bugs with NPU. On the other hand, is it possible to add NPU unit tests to the workflow if we provide an NPU CI server?

@conglongli
Copy link
Contributor

@RUAN-ZX Since we don't have the hardware, we will only review/approve based on whether your PR pass all our unit tests (it already failed the formatting test). And we would have to redirect all related future bug reports to you.

Of course, we'll fix all the bugs with NPU. On the other hand, is it possible to add NPU unit tests to the workflow if we provide an NPU CI server?

@CurryRice233 please contact @tjruwase for this kind of requests.

@conglongli conglongli enabled auto-merge December 8, 2023 04:31
@conglongli conglongli added this pull request to the merge queue Dec 8, 2023
auto-merge was automatically disabled December 8, 2023 06:32

Head branch was pushed to by a user without write access

@RUAN-ZX RUAN-ZX requested a review from conglongli December 8, 2023 06:39
@conglongli conglongli removed this pull request from the merge queue due to a manual request Dec 8, 2023
@RUAN-ZX
Copy link
Contributor Author

RUAN-ZX commented Dec 8, 2023

@conglongli, could you review and merge a latest commit? Sorry for this inconvenience and thank you :)

@conglongli
Copy link
Contributor

conglongli commented Dec 8, 2023

@conglongli, could you review and merge a latest commit? Sorry for this inconvenience and thank you :)

@RUAN-ZX In the future please avoid adding last minute commit which will break our auto-merge mechanism and keep delaying the PR merge.

@conglongli conglongli enabled auto-merge December 8, 2023 20:58
@conglongli conglongli added this pull request to the merge queue Dec 8, 2023
Merged via the queue into deepspeedai:master with commit 2ce6bf8 Dec 9, 2023
mauryaavinash95 pushed a commit to mauryaavinash95/DeepSpeed that referenced this pull request Feb 17, 2024
…ai#4733)

To support NPU devices fulfilling feature requirements like 1-bit Adam,
1-bit Lamb, 0/1 Adam, I add HcclBackend and its corresponding import
logics.
See what we have already done in deepspeedai#4567 .

---------

Co-authored-by: ryan <ruanzhixiang1@huawei.com>
Co-authored-by: Conglong Li <conglong.li@gmail.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants