Skip to content

Allow triton==3.0.x for fp_quantizer - #6447

Merged
loadams merged 3 commits into
deepspeedai:masterfrom
abacusai:master
Aug 28, 2024
Merged

Allow triton==3.0.x for fp_quantizer#6447
loadams merged 3 commits into
deepspeedai:masterfrom
abacusai:master

Conversation

@siddartha-RE

Copy link
Copy Markdown
Contributor

Tested with triton==3.0.x and the kernel tests pass so adding as an allowed version.

Triton 2.3.x is not well supported on arm64. Triton 3.0.0 is supported on arm64 and it appears the fp8 kernel works fine with triton==3.0.0 so this simplifies usage on arm hosts (GH200).

Tested with triton==3.0.x and the kernel tests pass so adding as an allowed version.
Comment thread op_builder/fp_quantizer.py Outdated
Comment thread op_builder/fp_quantizer.py Outdated
Comment thread op_builder/fp_quantizer.py Outdated
@siddartha-RE

Copy link
Copy Markdown
Contributor Author

@microsoft-github-policy-service agree [company="{Abacus.AI}"]

@siddartha-RE

Copy link
Copy Markdown
Contributor Author

@microsoft-github-policy-service agree company="Abacus.AI"

@loadams
loadams requested a review from jeffra August 27, 2024 21:23

@jeffra jeffra left a comment

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I ran tests on an H100 with triton 3.0.0 and all looks good to me. Thank you for the PR @siddartha-RE!

@loadams
loadams added this pull request to the merge queue Aug 28, 2024
Merged via the queue into deepspeedai:master with commit 4864991 Aug 28, 2024
mauryaavinash95 pushed a commit to DataStates/DeepSpeed that referenced this pull request Mar 20, 2025
Tested with triton==3.0.x and the kernel tests pass so adding as an
allowed version.

Triton 2.3.x is not well supported on arm64. Triton 3.0.0 is supported
on arm64 and it appears the fp8 kernel works fine with triton==3.0.0 so
this simplifies usage on arm hosts (GH200).
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