Skip to content

Arm backend: Keep DeepSeek FP8 on xlarge path - #21612

Merged
bdemirb merged 2 commits into
pytorch:mainfrom
bdemirb:baris_mletorch-2337_fp8_layer_test_DeepSeek_R1_Distill_Qwen_1.5B_reland
Aug 7, 2026
Merged

Arm backend: Keep DeepSeek FP8 on xlarge path#21612
bdemirb merged 2 commits into
pytorch:mainfrom
bdemirb:baris_mletorch-2337_fp8_layer_test_DeepSeek_R1_Distill_Qwen_1.5B_reland

Conversation

@bdemirb

@bdemirb bdemirb commented Aug 6, 2026

Copy link
Copy Markdown
Collaborator

Keep the DeepSeek-R1-Distill-Qwen-1.5B FP8 layer tests on the existing generic xlarge TOSA and VKML model suites.

This drops the dedicated DeepSeek job wiring added in the previous patchset and follows the existing Arm backend model for memory-heavy coverage: smaller cases stay in the normal shard and heavy cases run through the serialised xlarge path.

That keeps the FP8 coverage in CI without introducing DeepSeek-specific test entrypoints or workflow jobs.

cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani

Keep the DeepSeek-R1-Distill-Qwen-1.5B FP8 layer tests
on the existing generic xlarge TOSA and VKML model suites.

This drops the dedicated DeepSeek job wiring added in the
previous patchset and follows the existing Arm backend model
for memory-heavy coverage: smaller cases stay in the normal
shard and heavy cases run through the serialized xlarge path.

That keeps the FP8 coverage in CI without introducing
DeepSeek-specific test entrypoints or workflow jobs.

Authored with Codex.

Signed-off-by: Baris Demir <baris.demir@arm.com>

Change-Id: I2337d551a1b3c6c2d23b09d697c7713de6d9a5c0
@bdemirb
bdemirb requested a review from digantdesai as a code owner August 6, 2026 07:54
@pytorch-bot

pytorch-bot Bot commented Aug 6, 2026

Copy link
Copy Markdown

🔗 Helpful Links

🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21612

Note: Links to docs will display an error until the docs builds have been completed.

❗ 1 Active SEVs

There are 1 currently active SEVs. If your PR is affected, please view them below:

❌ 7 New Failures, 1 Unrelated Failure

As of commit 1750c29 with merge base 8702327 (image):

NEW FAILURES - The following jobs have failed:

FLAKY - The following job failed but was likely due to flakiness present on trunk:

This comment was automatically generated by Dr. CI and updates every 15 minutes.

@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. label Aug 6, 2026
@github-actions github-actions Bot added ciflow/trunk module: arm Issues related to arm backend labels Aug 6, 2026
@bdemirb

bdemirb commented Aug 6, 2026

Copy link
Copy Markdown
Collaborator Author

@pytorchbot label "partner: arm"

@pytorch-bot pytorch-bot Bot added the partner: arm For backend delegation, kernels, demo, etc. from the 3rd-party partner, Arm label Aug 6, 2026
@bdemirb

bdemirb commented Aug 6, 2026

Copy link
Copy Markdown
Collaborator Author

@pytorchbot label "release notes: arm"

@pytorch-bot pytorch-bot Bot added the release notes: arm Changes to the ARM backend delegate label Aug 6, 2026

@zingo zingo 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.

Just a note: test_pytest_models_*_xlarge don't seem to be run on GitHub is that on purpos?
e.g. is the model too large?

@bdemirb
bdemirb merged commit 70594d3 into pytorch:main Aug 7, 2026
486 of 494 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ciflow/trunk CLA Signed This label is managed by the Facebook bot. Authors need to sign the CLA before a PR can be reviewed. module: arm Issues related to arm backend partner: arm For backend delegation, kernels, demo, etc. from the 3rd-party partner, Arm release notes: arm Changes to the ARM backend delegate

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants