From 143e1f46d0bc764713accb94b3ea38267c9881ad Mon Sep 17 00:00:00 2001
From: MengLong Chen <71744434+dragondream-chen@users.noreply.github.com>
Date: Mon, 1 Dec 2025 20:44:11 +0800
Subject: [PATCH] [Feat] shared expert dp for deepseek_mtp (#3811)
### What this PR does / why we need it?
Support shared expert DP for deepseek_mtp feature.
`shared_expert_dp` requires `SP==True`, with corresponding parameter
restrictions.
Previously, due to the coupling between `shared_expert_dp` and torchair,
and the removal of `deepseek_mtp` in vllm_ascend, shared expert dp of
deepseek_mtp was temporarily removed.
Currently, by performing the `reduce_scatter` on the input of
deepssek_mtp in `mtp_proposer.py`, we ensure that it matches the
dimensions of `input_embedding`, and then perform the `all_gather` on
the output of mtp.
### How was this patch tested?
baseline:
enable shared_expert_dp and multistream_overlap_shared_expert:
TPOT: 48ms -> 45.4ms
Average TPS per rank: 117.6 -> 126.1
- vLLM version: v0.11.2
- vLLM main: https://github.com/vllm-project/vllm/commit/v0.11.2
---------
Signed-off-by: chenmenglong
Signed-off-by: zengran
Co-authored-by: zengran
---
tests/e2e/multicard/test_shared_expert_dp.py | 93 ++++++++++++++++++++
tests/ut/ops/test_layernorm.py | 7 +-
vllm_ascend/ascend_config.py | 4 +
vllm_ascend/attention/mla_v1.py | 2 +
vllm_ascend/ops/layernorm.py | 1 +
vllm_ascend/ops/register_custom_ops.py | 28 ++++++
vllm_ascend/platform.py | 4 +-
vllm_ascend/spec_decode/mtp_proposer.py | 55 +++++++++---
vllm_ascend/utils.py | 8 +-
9 files changed, 185 insertions(+), 17 deletions(-)
create mode 100644 tests/e2e/multicard/test_shared_expert_dp.py
diff --git a/tests/e2e/multicard/test_shared_expert_dp.py b/tests/e2e/multicard/test_shared_expert_dp.py
new file mode 100644
index 00000000..867d3ab6
--- /dev/null
+++ b/tests/e2e/multicard/test_shared_expert_dp.py
@@ -0,0 +1,93 @@
+import os
+
+import pytest
+from vllm import SamplingParams
+
+from tests.e2e.conftest import VllmRunner
+from tests.e2e.model_utils import check_outputs_equal
+
+MODELS = [
+ "vllm-ascend/DeepSeek-V2-Lite",
+]
+os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn"
+
+
+@pytest.mark.parametrize("model", MODELS)
+def test_models_with_enable_shared_expert_dp(model: str) -> None:
+
+ if 'HCCL_OP_EXPANSION_MODE' in os.environ:
+ del os.environ['HCCL_OP_EXPANSION_MODE']
+
+ prompts = [
+ "Hello, my name is", "The capital of the United States is",
+ "The capital of France is", "The future of AI is"
+ ]
+ sampling_params = SamplingParams(max_tokens=32, temperature=0.0)
+
+ with VllmRunner(
+ model,
+ max_model_len=1024,
+ enforce_eager=True,
+ tensor_parallel_size=2,
+ enable_expert_parallel=True,
+ ) as runner:
+ vllm_eager_outputs = runner.model.generate(prompts, sampling_params)
+
+ os.environ["VLLM_ASCEND_ENABLE_FLASHCOMM1"] = "1"
+ with VllmRunner(
+ model,
+ max_model_len=1024,
+ enforce_eager=True,
+ tensor_parallel_size=2,
+ enable_expert_parallel=True,
+ additional_config={
+ "enable_shared_expert_dp": True,
+ },
+ ) as runner:
+ shared_expert_dp_eager_outputs = runner.model.generate(
+ prompts, sampling_params)
+
+ with VllmRunner(
+ model,
+ max_model_len=1024,
+ tensor_parallel_size=2,
+ enforce_eager=False,
+ compilation_config={
+ "cudagraph_capture_sizes": [1, 4, 8, 16],
+ "cudagraph_mode": "FULL_DECODE_ONLY",
+ },
+ additional_config={
+ "enable_shared_expert_dp": True,
+ },
+ ) as runner:
+ shared_expert_dp_aclgraph_outputs = runner.model.generate(
+ prompts, sampling_params)
+
+ vllm_eager_outputs_list = []
+ for output in vllm_eager_outputs:
+ vllm_eager_outputs_list.append(
+ (output.outputs[0].index, output.outputs[0].text))
+
+ shared_expert_dp_eager_outputs_list = []
+ for output in shared_expert_dp_eager_outputs:
+ shared_expert_dp_eager_outputs_list.append(
+ (output.outputs[0].index, output.outputs[0].text))
+
+ shared_expert_dp_aclgraph_outputs_list = []
+ for output in shared_expert_dp_aclgraph_outputs:
+ shared_expert_dp_aclgraph_outputs_list.append(
+ (output.outputs[0].index, output.outputs[0].text))
+
+ check_outputs_equal(
+ outputs_0_lst=vllm_eager_outputs_list,
+ outputs_1_lst=shared_expert_dp_eager_outputs_list,
+ name_0="vllm_eager_outputs",
+ name_1="shared_expert_dp_eager_outputs",
+ )
+
+ check_outputs_equal(
+ outputs_0_lst=vllm_eager_outputs_list,
+ outputs_1_lst=shared_expert_dp_aclgraph_outputs_list,
+ name_0="vllm_eager_outputs",
+ name_1="shared_expert_dp_aclgraph_outputs",
+ )
diff --git a/tests/ut/ops/test_layernorm.py b/tests/ut/ops/test_layernorm.py
index 314775f8..77af2649 100644
--- a/tests/ut/ops/test_layernorm.py
+++ b/tests/ut/ops/test_layernorm.py
@@ -1,4 +1,5 @@
import unittest
+from unittest.mock import patch
import pytest
import torch
@@ -42,7 +43,9 @@ class TestAscendRMSNorm(PytestBase):
# Test case for the most common and basic scenario
@pytest.mark.parametrize(
"residual", [None, torch.randn(4, 8, dtype=torch.float16)])
- def test_forward_oot_basic(self, residual):
+ @patch("torch.ops.vllm.maybe_chunk_residual")
+ def test_forward_oot_basic(self, mock_maybe_chunk_residual, residual):
+ mock_maybe_chunk_residual.side_effect = lambda x, residual: residual
layer = RMSNorm(hidden_size=8, eps=1e-05)
x = torch.randn(4, 8, dtype=torch.float16)
if residual is not None:
@@ -107,6 +110,8 @@ class TestAscendRMSNorm(PytestBase):
mock_forward_context.num_hidden_layers = num_hidden_layers
mock_forward_context.fusion_linear = "gate_up_dense"
mock_forward_context.weight_prefetch_method = None
+ mocker.patch("torch.ops.vllm.maybe_chunk_residual",
+ lambda x, residual: residual)
# Ensure fusion and layer_idx increment are handled correctly
x = torch.randn(4, 8, dtype=torch.float16)
diff --git a/vllm_ascend/ascend_config.py b/vllm_ascend/ascend_config.py
index 16d16a4d..115dbef1 100644
--- a/vllm_ascend/ascend_config.py
+++ b/vllm_ascend/ascend_config.py
@@ -72,6 +72,10 @@ class AscendConfig:
self.enable_shared_expert_dp = additional_config.get(
"enable_shared_expert_dp", False
) and not self.torchair_graph_config.enabled and vllm_config.parallel_config.enable_expert_parallel
+ if self.enable_shared_expert_dp:
+ from vllm_ascend.utils import enable_sp
+ assert enable_sp(vllm_config=vllm_config,
+ enable_shared_expert_dp=True)
self.multistream_overlap_shared_expert = additional_config.get(
"multistream_overlap_shared_expert", False)
self.recompute_scheduler_enable = additional_config.get(
diff --git a/vllm_ascend/attention/mla_v1.py b/vllm_ascend/attention/mla_v1.py
index 19c8025b..5d341d03 100644
--- a/vllm_ascend/attention/mla_v1.py
+++ b/vllm_ascend/attention/mla_v1.py
@@ -1677,6 +1677,8 @@ class AscendMLAImpl(MLAAttentionImpl):
forward_context = get_forward_context()
if (self.enable_mlapo and
(attn_metadata is None or not forward_context.with_prefill)):
+ hidden_states = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(
+ hidden_states.contiguous(), need_gather_q_kv)
decode_preprocess_res, prefill_preprocess_res = self._mla_decode_preprocess(
hidden_states, kv_cache, attn_metadata)
else:
diff --git a/vllm_ascend/ops/layernorm.py b/vllm_ascend/ops/layernorm.py
index 8c395b54..da5051c0 100644
--- a/vllm_ascend/ops/layernorm.py
+++ b/vllm_ascend/ops/layernorm.py
@@ -110,6 +110,7 @@ class AscendRMSNorm(RMSNorm):
import torch_npu
if residual is not None:
+ residual = torch.ops.vllm.maybe_chunk_residual(x, residual)
assert x.size(0) == residual.size(0)
x, residual = _addrmsnorm_forward_oot(
self, x, residual, self.next_need_quant_fusion_linear,
diff --git a/vllm_ascend/ops/register_custom_ops.py b/vllm_ascend/ops/register_custom_ops.py
index bb16bc00..03bea554 100644
--- a/vllm_ascend/ops/register_custom_ops.py
+++ b/vllm_ascend/ops/register_custom_ops.py
@@ -2,6 +2,7 @@ import torch
import torch.nn.functional as F
import torch_npu
from vllm.distributed import (get_dp_group, get_ep_group,
+ get_tensor_model_parallel_rank,
get_tensor_model_parallel_world_size,
tensor_model_parallel_all_gather,
tensor_model_parallel_all_reduce,
@@ -15,6 +16,27 @@ from vllm_ascend.ops.weight_prefetch import maybe_npu_prefetch
from vllm_ascend.utils import npu_stream_switch, prefetch_stream
+def _maybe_chunk_residual_impl(x: torch.Tensor,
+ residual: torch.Tensor) -> torch.Tensor:
+ try:
+ forward_context = get_forward_context()
+ except AssertionError:
+ return residual
+
+ if x.size(0) != residual.size(0):
+ sp_enabled = forward_context.sp_enabled
+ assert sp_enabled is True, ("Currently, this situation only occurs "
+ "when sp is enabled")
+ pad_size = forward_context.pad_size
+ if pad_size > 0:
+ residual = F.pad(residual, (0, 0, 0, pad_size))
+ tp_size = get_tensor_model_parallel_world_size()
+ tp_rank = get_tensor_model_parallel_rank()
+ residual = torch.chunk(residual, tp_size, dim=0)[tp_rank]
+
+ return residual
+
+
def _maybe_all_gather_and_maybe_unpad_impl(
x: torch.Tensor,
label: bool,
@@ -259,6 +281,12 @@ def _matmul_and_reduce_impl_fake(input_parallel: torch.Tensor,
return output
+direct_register_custom_op(op_name="maybe_chunk_residual",
+ op_func=_maybe_chunk_residual_impl,
+ fake_impl=lambda x, residual: x,
+ mutates_args=[],
+ dispatch_key="PrivateUse1")
+
direct_register_custom_op(op_name="maybe_all_gather_and_maybe_unpad",
op_func=_maybe_all_gather_and_maybe_unpad_impl,
fake_impl=_maybe_all_gather_and_maybe_unpad_fake,
diff --git a/vllm_ascend/platform.py b/vllm_ascend/platform.py
index 7cc84fc6..5ff66926 100644
--- a/vllm_ascend/platform.py
+++ b/vllm_ascend/platform.py
@@ -283,7 +283,7 @@ class NPUPlatform(Platform):
if parallel_config and parallel_config.worker_cls == "auto":
# TODO: this is a tricky way to disable `use_sequence_parallel_moe` in vllm.
parallel_config.all2all_backend = "flashinfer_all2allv"
- if ascend_config.torchair_graph_config.enabled or ascend_config.enable_shared_expert_dp:
+ if ascend_config.torchair_graph_config.enabled:
parallel_config.worker_cls = "vllm_ascend.torchair.torchair_worker.NPUTorchairWorker"
else:
parallel_config.worker_cls = "vllm_ascend.worker.worker_v1.NPUWorker"
@@ -379,8 +379,6 @@ class NPUPlatform(Platform):
ascend_config = get_ascend_config()
if use_mla and ascend_config.enable_shared_expert_dp:
- if use_mla and not use_sparse:
- return "vllm_ascend.torchair.torchair_mla.AscendMLATorchairBackend"
if use_mla and use_sparse:
return "vllm_ascend.torchair.torchair_sfa.AscendSFATorchairBackend"
diff --git a/vllm_ascend/spec_decode/mtp_proposer.py b/vllm_ascend/spec_decode/mtp_proposer.py
index 15b2b4cf..cacc2bdf 100644
--- a/vllm_ascend/spec_decode/mtp_proposer.py
+++ b/vllm_ascend/spec_decode/mtp_proposer.py
@@ -32,7 +32,8 @@ from vllm_ascend.compilation.acl_graph import (ACLGraphWrapper,
update_mla_attn_params)
from vllm_ascend.spec_decode.interface import Proposer, SpecDcodeType
from vllm_ascend.utils import (ProfileExecuteDuration, lmhead_tp_enable,
- prefill_context_parallel_enable)
+ prefill_context_parallel_enable,
+ shared_expert_dp_enabled)
if prefill_context_parallel_enable():
from vllm.distributed import get_pcp_group
@@ -94,6 +95,7 @@ class MtpProposer(Proposer):
# the draft model's hidden size can be different from the target model's
# hidden size (e.g., Llama 3.3 70B).
self.hidden_size = self.draft_model_config.get_hidden_size()
+ self.enable_shared_expert_dp = shared_expert_dp_enabled()
self.pcp_size = self.runner.pcp_size
self.dcp_size = self.runner.dcp_size
@@ -286,6 +288,12 @@ class MtpProposer(Proposer):
aclgraph_runtime_mode=aclgraph_runtime_mode,
batch_descriptor=batch_descriptor,
is_mtp_model=True):
+ if self.enable_shared_expert_dp:
+ positions = positions.unsqueeze(-1)
+ positions = torch.ops.vllm.maybe_pad_and_reduce(positions)
+ positions = positions.squeeze(-1)
+ previous_hidden_states = torch.ops.vllm.maybe_pad_and_reduce(
+ previous_hidden_states)
self.model(input_ids=input_ids,
positions=positions,
hidden_states=previous_hidden_states)
@@ -294,9 +302,13 @@ class MtpProposer(Proposer):
not forward_context.capturing:
if self.vllm_config.model_config.use_mla:
update_mla_attn_params(
- self.update_stream, forward_context,
- positions.shape[0],
+ self.update_stream, forward_context, num_tokens,
self.vllm_config.speculative_config)
+ if self.enable_shared_expert_dp:
+ positions = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(
+ positions, True)
+ previous_hidden_states = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(
+ previous_hidden_states, True)
dummy_compute_logits(previous_hidden_states)
if with_prefill:
break
@@ -675,7 +687,8 @@ class MtpProposer(Proposer):
moe_comm_type = self.runner._select_moe_comm_method(num_input_tokens)
- if scheduler_output:
+ # Enable shared_expert_dp and MTP FULL graph may cause accuracy issues.
+ if scheduler_output and not self.enable_shared_expert_dp:
max_query_len = common_attn_metadata.max_query_len
uniform_decode = (max_query_len in list(
range(1, self.num_speculative_tokens +
@@ -725,11 +738,22 @@ class MtpProposer(Proposer):
with ProfileExecuteDuration().capture_async('mtp_forward'):
model_kwargs = {}
model_kwargs["attn_metadata"] = attn_metadata
+ input_ids = self.input_ids[:num_input_tokens]
+ positions = self.positions[:num_input_tokens]
+ hidden_states = self.hidden_states[:num_input_tokens]
- hidden_states = self.model(
- input_ids=self.input_ids[:num_input_tokens],
- positions=self.positions[:num_input_tokens],
- hidden_states=self.hidden_states[:num_input_tokens])
+ if self.enable_shared_expert_dp:
+ # positions [N] -> [N, 1] for padding
+ positions = positions.unsqueeze(-1)
+ positions = torch.ops.vllm.maybe_pad_and_reduce(
+ positions)
+ positions = positions.squeeze(-1)
+ hidden_states = torch.ops.vllm.maybe_pad_and_reduce(
+ hidden_states)
+
+ hidden_states = self.model(input_ids=input_ids,
+ positions=positions,
+ hidden_states=hidden_states)
forward_context = get_forward_context()
if forward_context.cudagraph_runtime_mode == CUDAGraphMode.FULL:
if self.vllm_config.model_config.use_mla:
@@ -738,6 +762,12 @@ class MtpProposer(Proposer):
num_input_tokens,
self.vllm_config.speculative_config)
+ if self.enable_shared_expert_dp:
+ hidden_states = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(
+ hidden_states.contiguous(), True)
+ positions = torch.ops.vllm.maybe_all_gather_and_maybe_unpad(
+ positions.contiguous(), True)
+
num_indices = last_token_indices.shape[0]
if lmhead_tp_enable():
if not self.runner.with_prefill:
@@ -805,20 +835,21 @@ class MtpProposer(Proposer):
batch_size,
attn_metadata_i.decode.actual_seq_lengths_q)
attn_metadata_i.decode.cos = builder.cos_cache[
- positions].unsqueeze(1).unsqueeze(2)
+ positions[:batch_size]].unsqueeze(1).unsqueeze(2)
attn_metadata_i.decode.sin = builder.sin_cache[
- positions].unsqueeze(1).unsqueeze(2)
+ positions[:batch_size]].unsqueeze(1).unsqueeze(2)
# NOTE(woosuk): We should handle the case where the draft model
# generates tokens beyond the max model length. Since it is complex
# to remove such requests from the batch, we keep them in the batch
# but adjust the position ids and slot mappings to avoid the
# out-of-range access during the model execution. The draft tokens
# generated with this adjustment should be ignored.
- exceeds_max_model_len = positions >= self.runner.model_config.max_model_len
+ exceeds_max_model_len = positions[:
+ batch_size] >= self.runner.model_config.max_model_len
# Mask out the position ids that exceed the max model length.
# Otherwise, we may get out-of-range error in RoPE.
clamped_positions = torch.where(exceeds_max_model_len, 0,
- positions)
+ positions[:batch_size])
# Increment the sequence lengths.
attn_metadata_i.seq_lens[:batch_size] += 1
# For the requests that exceed the max model length, we set the
diff --git a/vllm_ascend/utils.py b/vllm_ascend/utils.py
index b41034a4..e9441e28 100644
--- a/vllm_ascend/utils.py
+++ b/vllm_ascend/utils.py
@@ -758,7 +758,7 @@ def dense_optim_enable() -> bool:
return envs_ascend.VLLM_ASCEND_ENABLE_DENSE_OPTIMIZE
-def enable_sp(vllm_config=None) -> bool:
+def enable_sp(vllm_config=None, enable_shared_expert_dp: bool = False) -> bool:
global _ENABLE_SP
if _ENABLE_SP is None:
if vllm_config is None:
@@ -772,6 +772,12 @@ def enable_sp(vllm_config=None) -> bool:
# We retain the env VLLM_ASCEND_ENABLE_FLASHCOMM here for backward compatibility.
or bool(int(os.getenv("VLLM_ASCEND_ENABLE_FLASHCOMM", '0'))))
+ if not _ENABLE_SP and enable_shared_expert_dp:
+ _ENABLE_SP = True
+ logger.info(
+ "shared_expert_dp requires enable_sp = True. has set enable_sp to True"
+ )
+
if not _ENABLE_SP:
return _ENABLE_SP