[Bugfix] [MoE] fix error in deepseek when using allgather (#3824)

### What this PR does / why we need it?
After refactoring vllm_ascend/models and FusedMoE, we are unable to pass
`gate` from deepseekv2.py to `AscendFusedMoE.forward`, which will result
in error when running deepseek v3/r1 with allgather.
Hence, this pr removes `gate` related computations from FusedMoE module
in eager/aclgraph mode.
### Does this PR introduce _any_ user-facing change?
`rm_router_logits` is deprecated in eager/aclgraph.
### How was this patch tested?
e2e & ut

- vLLM version: v0.11.0rc3
- vLLM main:
https://github.com/vllm-project/vllm/commit/releases/v0.11.1

Signed-off-by: Pr0Wh1teGivee <calvin_zhu0210@outlook.com>
This commit is contained in:
weichen
2025-10-29 14:51:39 +08:00
committed by GitHub
parent 900086fdc6
commit 0d1859af08
7 changed files with 56 additions and 85 deletions

View File

@@ -67,7 +67,7 @@ class TestMoECommMethod(TestBase):
# Verify prepare was called with correct arguments
mock_pf_instance.prepare.assert_called_once_with(
hidden_states, router_logits, False, False, None)
hidden_states, router_logits, False, False)
# Test finalize method
comm_impl.finalize(h_out,
@@ -115,7 +115,7 @@ class TestMoECommMethod(TestBase):
# Verify prepare was called with correct arguments
mock_pf_instance.prepare.assert_called_once_with(
hidden_states, router_logits, False, False, None)
hidden_states, router_logits, False, False)
# Test finalize method
comm_impl.finalize(h_out,
@@ -165,7 +165,7 @@ class TestMoECommMethod(TestBase):
# Verify prepare was called with correct arguments
mock_pf_instance.prepare.assert_called_once_with(
hidden_states, router_logits, False, False, None)
hidden_states, router_logits, False, False)
@patch("vllm_ascend.ops.fused_moe.moe_comm_method.get_current_vllm_config")
@patch("vllm_ascend.ops.fused_moe.moe_comm_method.get_forward_context")