Skip to content

[Bugfix] Fix GPT-OSS AR+NORM fusion #28841

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

Merged
ProExpertProg merged 8 commits into from
Nov 25, 2025
Merged

[Bugfix] Fix GPT-OSS AR+NORM fusion #28841

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
8 commits
Select commit Hold shift + click to select a range
3318069
fix GPT-OSS AR+NORM fusion
elvischenv Nov 13, 2025
1cdafbf
add gpt-oss-20b e2e fusion test
elvischenv Nov 18, 2025
c6b72cd
fix memory contiguous issue correctly
elvischenv Nov 18, 2025
54ab93f
add fused_moe to test dependencies
elvischenv Nov 19, 2025
f2bf032
Merge branch 'main' into elvischenv/fix-gpt-oss-ar-norm-fusion
DarkLight1337 Nov 21, 2025
d05fab1
Merge branch 'main' into elvischenv/fix-gpt-oss-ar-norm-fusion
elvischenv Nov 21, 2025
3b85853
Merge branch 'main' into elvischenv/fix-gpt-oss-ar-norm-fusion
mgoin Nov 21, 2025
9423e90
Merge branch 'main' into elvischenv/fix-gpt-oss-ar-norm-fusion
elvischenv Nov 25, 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
1 change: 1 addition & 0 deletions .buildkite/test-pipeline.yaml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -970,6 +970,7 @@ steps:
- vllm/model_executor/layers/layernorm.py
- vllm/model_executor/layers/activation.py
- vllm/model_executor/layers/quantization/input_quant_fp8.py
- vllm/model_executor/layers/fused_moe/layer.py
- tests/compile/test_fusion_attn.py
- tests/compile/test_silu_mul_quant_fusion.py
- tests/compile/distributed/test_fusion_all_reduce.py
Expand Down
11 changes: 11 additions & 0 deletions tests/compile/distributed/test_fusions_e2e.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -111,6 +111,17 @@ class ModelBackendTestCase(NamedTuple):
async_tp=96, # MLP is MoE, half the fusions of dense
),
),
ModelBackendTestCase(
model_name="openai/gpt-oss-20b",
model_kwargs=dict(max_model_len=1024, kv_cache_dtype="fp8"),
backend=AttentionBackendEnum.FLASHINFER,
matches=Matches(
attention_fusion=0,
allreduce_fusion=49,
sequence_parallel=49,
async_tp=48,
),
),
Comment on lines +114 to +124
Copy link
Contributor Author

@elvischenv elvischenv Nov 18, 2025

Choose a reason for hiding this comment

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

@ProExpertProg @mgoin Added the 20b e2e fusion test.
Also tested on main and got expected failure:

>   assert int(log_matches[0]) == matches.allreduce_fusion
      ^^^^^^^^^^^^^^^^^
E   AssertionError: assert 25 == 49
E    +  where 25 = int('25')
E    +  and   49 = Matches(attention_fusion=0, allreduce_fusion=49, sequence_parallel=49, async_tp=48).allreduce_fusion

]

elif current_platform.is_rocm():
Expand Down
2 changes: 1 addition & 1 deletion vllm/distributed/device_communicators/symm_mem.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -131,7 +131,7 @@ def all_reduce(
return None
if out is None:
out = torch.empty_like(inp)
self.buffer[: inp.numel()].copy_(inp.view(-1))
self.buffer[: inp.numel()].copy_(inp.reshape(-1))

# Determine which algorithm to use
use_multimem = False
Expand Down
17 changes: 11 additions & 6 deletions vllm/model_executor/layers/fused_moe/layer.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1690,6 +1690,10 @@ def forward_native(
)

def reduce_output(states: torch.Tensor) -> torch.Tensor:
# Slice before all_reduce to enable possible fusion
if self.hidden_size != og_hidden_states:
states = states[..., :og_hidden_states]

if (
not self.is_sequence_parallel
and not self.use_dp_chunking
Expand All @@ -1712,11 +1716,12 @@ def reduce_output(states: torch.Tensor) -> torch.Tensor:
if self.zero_expert_num is not None and self.zero_expert_num > 0:
assert isinstance(fused_output, tuple)
fused_output, zero_expert_result = fused_output
return (reduce_output(fused_output) + zero_expert_result)[
..., :og_hidden_states
]
return (
reduce_output(fused_output)
+ zero_expert_result[..., :og_hidden_states]
)
else:
return reduce_output(fused_output)[..., :og_hidden_states]
return reduce_output(fused_output)
else:
if current_platform.is_tpu():
# TODO: Once the OOM issue for the TPU backend is resolved, we
Expand All @@ -1729,8 +1734,8 @@ def reduce_output(states: torch.Tensor) -> torch.Tensor:
hidden_states, router_logits, self.layer_name
)
return (
reduce_output(shared_output)[..., :og_hidden_states],
reduce_output(fused_output)[..., :og_hidden_states],
reduce_output(shared_output),
reduce_output(fused_output),
)

def forward_cuda(
Expand Down