Skip to content

[nvbugs/5368410][fix] Disable moe allreduce for multi node #5918

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
yizhang-nv merged 1 commit into from
Jul 14, 2025
Merged

[nvbugs/5368410][fix] Disable moe allreduce for multi node #5918

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
7 changes: 4 additions & 3 deletions jenkins/L0_Test.groovy
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1825,9 +1825,10 @@ def launchTestJobs(pipeline, testFilter, dockerNode=null)

multiNodesSBSAConfigs = [
// Each stage test 1 testcase with 8 GPUs and 2 nodes.
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-1": ["gb200-multi-node", "l0_gb200_multi_nodes", 1, 3, 8, 2],
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-2": ["gb200-multi-node", "l0_gb200_multi_nodes", 2, 3, 8, 2],
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-3": ["gb200-multi-node", "l0_gb200_multi_nodes", 3, 3, 8, 2],
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-1": ["gb200-multi-node", "l0_gb200_multi_nodes", 1, 4, 8, 2],
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-2": ["gb200-multi-node", "l0_gb200_multi_nodes", 2, 4, 8, 2],
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-3": ["gb200-multi-node", "l0_gb200_multi_nodes", 3, 4, 8, 2],
"GB200-8_GPUs-2_Nodes-PyTorch-[Post-Merge]-4": ["gb200-multi-node", "l0_gb200_multi_nodes", 4, 4, 8, 2],
]
fullSet += multiNodesSBSAConfigs.keySet()

Expand Down
12 changes: 7 additions & 5 deletions tensorrt_llm/_torch/models/modeling_deepseekv3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@
from tqdm import tqdm
from transformers import PretrainedConfig

from tensorrt_llm._ipc_utils import can_access_peer
from tensorrt_llm._utils import get_sm_version
from tensorrt_llm.functional import PositionEmbeddingType
from tensorrt_llm.llmapi.utils import enable_llm_debug
Expand Down Expand Up @@ -604,6 +605,7 @@ def __init__(self, model_config: ModelConfig[PretrainedConfig],
self.enable_attention_dp = mapping.enable_attention_dp

self.mlp_tp_size = mapping.tp_size
self.is_p2p_supported = can_access_peer(mapping)

self.fusion_config = EagerFusionConfig()
self.enable_fusion = os.environ.get(
Expand Down Expand Up @@ -843,11 +845,11 @@ def _run_MoE(hidden_states, hidden_states_fp4, do_finalize):
hidden_states, residual)

# Note: this fusion pattern is only supported for TRTLLM-nvfp4 backend now
do_finalize = not (hidden_states.shape[0]
<= self.moe_allreduce.max_token
and self.fusion_config.POST_MOE_FUSION
and self.model_config.moe_backend == 'TRTLLM'
and self.mlp.experts.has_nvfp4)
do_finalize = not (
hidden_states.shape[0] <= self.moe_allreduce.max_token
and self.fusion_config.POST_MOE_FUSION
and self.model_config.moe_backend == 'TRTLLM'
and self.mlp.experts.has_nvfp4 and self.is_p2p_supported)

hidden_states = _run_MoE(hidden_states,
hidden_states_fp4=None,
Expand Down
1 change: 1 addition & 0 deletions tests/integration/test_lists/test-db/l0_gb200_multi_nodes.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,4 +15,5 @@ l0_gb200_multi_nodes:
tests:
- accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_8gpus[latency] TIMEOUT (180)
- accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_8gpus[throughput_tp8] TIMEOUT (180)
- accuracy/test_llm_api_pytorch.py::TestDeepSeekR1::test_nvfp4_8gpus[latency_trtllmgen] TIMEOUT (180)
- accuracy/test_llm_api_pytorch.py::TestQwen3_235B_A22B::test_nvfp4[latency] TIMEOUT (180)