Skip to content

[PyTorch] Error out if constructing LayerNormLinear with row tensor parallelism #2688

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
timmoon10 wants to merge 3 commits into
base: main
Choose a base branch
from
+4 −4
Open

[PyTorch] Error out if constructing LayerNormLinear with row tensor parallelism #2688

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
0f7674f
Error out if constructing LayerNormLinear with row tensor parallelism
timmoon10 Feb 17, 2026
0e225c1
Disable Userbuffers test for row-TP LayerNormLinear
timmoon10 Feb 20, 2026
1ef2ae0
Merge branch 'main' into tmoon/row-tp-layernorm-linear
timmoon10 Feb 20, 2026
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
4 changes: 0 additions & 4 deletions tests/pytorch/distributed/test_comm_gemm_overlap.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -210,7 +210,6 @@ def test_bulk_overlaps(comm_type, quantization, connections):
(te.Linear.__name__, "row", False),
(te.Linear.__name__, "column", False),
(te.Linear.__name__, "column", True),
(te.LayerNormLinear.__name__, "row", False),
(te.LayerNormLinear.__name__, "column", False),
(te.LayerNormLinear.__name__, "column", True),
]
Expand All @@ -225,7 +224,6 @@ def test_bulk_overlaps(comm_type, quantization, connections):
f" {te.Linear.__name__} - ROW-PARALLEL ",
f" {te.Linear.__name__} - COL-PARALLEL - BULK DGRAD/WGRAD ",
f" {te.Linear.__name__} - COL-PARLALEL - DGRAD+RS ",
f" {te.LayerNormLinear.__name__} - ROW-PARALLEL ",
f" {te.LayerNormLinear.__name__} - COL-PARALLEL - BULK DGRAD/WGRAD ",
f" {te.LayerNormLinear.__name__} - COL-PARALLEL - DGRAD+RS ",
]
Expand Down Expand Up @@ -254,7 +252,6 @@ def test_layers_with_overlap_bf16(layer_type, linear_parallel_mode, overlap_rs_d
(te.Linear.__name__, "row", False),
(te.Linear.__name__, "column", False),
(te.Linear.__name__, "column", True),
(te.LayerNormLinear.__name__, "row", False),
(te.LayerNormLinear.__name__, "column", False),
(te.LayerNormLinear.__name__, "column", True),
]
Expand All @@ -269,7 +266,6 @@ def test_layers_with_overlap_bf16(layer_type, linear_parallel_mode, overlap_rs_d
f"{te.Linear.__name__}-row_tensor_parallel",
f"{te.Linear.__name__}-col_tensor_parallel-BULK DGRAD/WGRAD",
f"{te.Linear.__name__}-col_tensor_parallel-DGRAD+RS",
f"{te.LayerNormLinear.__name__}-row_tensor_parallel",
f"{te.LayerNormLinear.__name__}-col_tensor_parallel-BULK DGRAD/WGRAD",
f"{te.LayerNormLinear.__name__}-col_tensor_parallel-DGRAD+RS",
]
Expand Down
4 changes: 4 additions & 0 deletions transformer_engine/pytorch/module/layernorm_linear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1193,6 +1193,10 @@ def __init__(
assert (
self.parallel_mode in GemmParallelModes
), f"parallel_mode {parallel_mode} not supported"
if self.parallel_mode == "row":
raise NotImplementedError(
"Normalization does not support tensor-parallel distribution."
)

if self.parallel_mode == "column":
self.out_features = divide(self.out_features, self.tp_size)
Expand Down
Loading