Skip to content

Make AC recompute test topology-robust#490

Merged
aditvenk merged 1 commit into
mainfrom
aditvenk/stack/2
Jun 15, 2026
Merged

Make AC recompute test topology-robust#490
aditvenk merged 1 commit into
mainfrom
aditvenk/stack/2

Conversation

@aditvenk

@aditvenk aditvenk commented Jun 12, 2026

Copy link
Copy Markdown
Contributor

Stacked PRs:


Make AC recompute test topology-robust

Accept MUST_RECOMPUTE nodes carrying AutoParallel's AC graph id, since those are AP-owned FSDP recompute tags and may include wait_tensor nodes depending on the selected sharding strategy.

Authored with Claude.

aditvenk added a commit that referenced this pull request Jun 12, 2026
Accept MUST_RECOMPUTE nodes carrying AutoParallel's AC graph id, since those are AP-owned FSDP recompute tags and may include wait_tensor nodes depending on the selected sharding strategy.

Authored with Claude.

stack-info: PR: #490, branch: aditvenk/stack/2
@meta-cla meta-cla Bot added the CLA Signed This label is managed by the Meta Open Source bot. label Jun 12, 2026
@aditvenk aditvenk marked this pull request as draft June 15, 2026 04:22
@aditvenk aditvenk changed the base branch from aditvenk/stack/1 to main June 15, 2026 04:22
@aditvenk aditvenk changed the base branch from main to aditvenk/stack/1 June 15, 2026 04:23
@aditvenk aditvenk marked this pull request as ready for review June 15, 2026 04:23

@fmassa fmassa left a comment

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

LGTM!

@@ -309,7 +309,10 @@ def input_fn():
def test_ac_joint_pass_marks_recomputable_nodes(device_mesh_1d):

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

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

Another alternative would be to wrap the tests with apply_cuda_patches

def apply_cuda_patches(func):

as this would make sure that tests assume the specific GPU arch.

Accept MUST_RECOMPUTE nodes carrying AutoParallel's AC graph id, since those are AP-owned FSDP recompute tags and may include wait_tensor nodes depending on the selected sharding strategy.

Authored with Claude.

stack-info: PR: #490, branch: aditvenk/stack/2
@aditvenk aditvenk marked this pull request as draft June 15, 2026 16:26
@aditvenk aditvenk changed the base branch from aditvenk/stack/1 to main June 15, 2026 16:26
@aditvenk aditvenk marked this pull request as ready for review June 15, 2026 16:28
@aditvenk aditvenk marked this pull request as draft June 15, 2026 16:37
@aditvenk aditvenk marked this pull request as ready for review June 15, 2026 16:37
@aditvenk aditvenk merged commit 1aff15c into main Jun 15, 2026
10 of 11 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

CLA Signed This label is managed by the Meta Open Source bot.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants