Arm backend: Add serialised xlarge VKML model suite - #21518
Open
bdemirb wants to merge 3 commits into
Open
Conversation
Relands the DeepSeek-R1-Distill-Qwen-1.5B layer tests that were reverted by pytorch#21048 because the TOSA model shard could run several checkpoint-shaped exports concurrently and OOM. The prerequisite CI change in pytorch#21492 added a serialized xlarge TOSA model suite while keeping the normal TOSA model shard parallelized. This reland marks the DeepSeek TOSA layer tests as xlarge so they are excluded from the normal shard and can be routed through the serialized xlarge suite. The tests use the checkpoint configuration from the Hugging Face model and the upstream Qwen2 layer implementations that back this distilled model. The covered layers include rotary embedding, rotary application, KV repetition, attention, RMSNorm, MLP, decoder layer, and final norm. Token embedding is excluded because the full checkpoint embedding allocation is too large for regular CI. Signed-off-by: Baris Demir <baris.demir@arm.com> Change-Id: Ia28581bbb4ffe070bc35af060fcceef2ac90084a
The DeepSeek-R1-Distill-Qwen layer tests were re-landed after the TOSA model shard OOM was addressed by a serialized xlarge TOSA suite. CommitlyExtended still showed memory pressure in the VKML model shard. The normal VKML model shard uses pytest-xdist auto parallelism, so xlarge VGF exports can also be resident at the same time. Add a serialized xlarge VKML model suite while keeping the normal VKML model shard parallelized. Mark the DeepSeek VGF layer tests as xlarge so they are excluded from normal VKML and can be routed through the serialized suite. Authored with Codex. Signed-off-by: Baris Demir <baris.demir@arm.com> Change-Id: I4b33f59a64ab2c40480e0f02c7415e999f4be7a2
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/executorch/21518
Note: Links to docs will display an error until the docs builds have been completed. ❌ 1 Cancelled JobAs of commit 7247430 with merge base 6683757 ( CANCELLED JOB - The following job was cancelled. Please retry:
This comment was automatically generated by Dr. CI and updates every 15 minutes. |
meta-cla
Bot
added
the
CLA Signed
bdemirb
changed the title
bdemirb
changed the title
Collaborator
Author
|
@pytorchbot label "partner: arm" |
pytorch-bot
Bot
added
the
partner: arm
Collaborator
Author
|
@pytorchbot label "release notes: arm" |
usamahz
self-requested a review
July 31, 2026 16:07
usamahz
approved these changes
Jul 31, 2026
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
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.
The DeepSeek-R1-Distill-Qwen layer tests were re-landed after the TOSA model shard OOM was addressed by a serialised xlarge TOSA suite.
CommitlyExtended still showed memory pressure in the VKML model shard. The normal VKML model shard uses pytest-xdist auto parallelism, so xlarge VGF exports can also be resident at the same time.
Add a serialised xlarge VKML model suite while keeping the normal VKML model shard parallelised. Mark the DeepSeek VGF layer tests as xlarge so they are excluded from normal VKML and can be routed through the serialised suite.
cc @digantdesai @freddan80 @per @zingo @oscarandersson8218 @mansnils @Sebastian-Larsson @robell @rascani