Skip to content

Enable AITER ASM distributed FA testing in jax/torch #363

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 4 commits into
base: dev
Choose a base branch
from
+4 −0
Open

Enable AITER ASM distributed FA testing in jax/torch #363

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
96a0158
Add AITER ASM enabled distributed FA testing in jax/torch
Micky774 Nov 4, 2025
2999f9f
Added V3 dist runs for older JAX versions
Micky774 Nov 5, 2025
3b072ff
Guard CK V3 tests based on backend
Micky774 Nov 10, 2025
72ae2c8
Merge branch 'dev' into zain/aiter-v3-dist
Micky774 Nov 12, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 3 additions & 0 deletions ci/jax.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -71,13 +71,16 @@ run_test_config_mgpu() {
*0.4.35*)
# Workaround for distributed tests hang with xla_flag
XLA_FLAGS="--xla_gpu_enable_nccl_comm_splitting=false" run 3 test_distributed_fused_attn.py -k 'not test_context_parallel_ring_attn'
XLA_FLAGS="--xla_gpu_enable_nccl_comm_splitting=false" NVTE_CK_USES_FWD_V3=1 NVTE_CK_USES_BWD_V3=1 run 3 test_distributed_fused_attn.py -k 'not test_context_parallel_ring_attn'
Copy link
Collaborator

@ipanfilo ipanfilo Nov 10, 2025

Choose a reason for hiding this comment

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

This will run it with AOTriton too

Copy link
Contributor Author

@Micky774 Micky774 Nov 10, 2025

Choose a reason for hiding this comment

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

Updated with a guard in the JAX ci script

Copy link
Collaborator

@ipanfilo ipanfilo Nov 14, 2025

Choose a reason for hiding this comment

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

With those changes env variables are not seen by run method - they are applied to test call only.
Using run_default_fa_lbl. All V3 calls should be labelled with "v3" to distinct them from regular test_distributed_fused_attn call


# Test ring attention with xla_flag --xla_experimental_ignore_channel_id only
XLA_FLAGS="--xla_experimental_ignore_channel_id" run_lbl "parallel_ring" 3 test_distributed_fused_attn.py -k test_context_parallel_ring_attn
XLA_FLAGS="--xla_experimental_ignore_channel_id" NVTE_CK_USES_FWD_V3=1 NVTE_CK_USES_BWD_V3=1 run_lbl "parallel_ring" 3 test_distributed_fused_attn.py -k test_context_parallel_ring_attn
;;
*)
# Workaround for distributed tests hang with xla_flag
XLA_FLAGS="--xla_gpu_enable_nccl_comm_splitting=false" run 3 test_distributed_fused_attn.py
XLA_FLAGS="--xla_gpu_enable_nccl_comm_splitting=false" NVTE_CK_USES_FWD_V3=1 NVTE_CK_USES_BWD_V3=1 run 3 test_distributed_fused_attn.py
;;
esac

Expand Down
1 change: 1 addition & 0 deletions ci/pytorch.sh
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -94,6 +94,7 @@ run_test_config_mgpu(){
run 3 distributed/test_numerics.py
run 3 distributed/test_torch_fsdp2.py
run 3 fused_attn/test_fused_attn_with_cp.py
NVTE_CK_USES_FWD_V3=1 NVTE_CK_USES_BWD_V3=1 run 3 fused_attn/test_fused_attn_with_cp.py
fi
}

Expand Down