[Feature] Expert Parallelism Load Balancer (EPLB) (#18343)

Signed-off-by: Bowen Wang <abmfy@icloud.com>

vllm/model_executor/models/deepseek_v2.py

@@ -23,7 +23,8 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 """Inference-only DeepseekV2/DeepseekV3 model."""
-from collections.abc import Iterable
+import typing
+from collections.abc import Callable, Iterable
 from typing import Any, Optional, Union
 
 import torch
@@ -32,8 +33,10 @@ from transformers import PretrainedConfig
 
 from vllm.attention import Attention
 from vllm.compilation.decorators import support_torch_compile
-from vllm.config import CacheConfig, ModelConfig, VllmConfig
-from vllm.distributed import get_pp_group, get_tensor_model_parallel_world_size
+from vllm.config import (CacheConfig, ModelConfig, VllmConfig,
+                         get_current_vllm_config)
+from vllm.distributed import (get_ep_group, get_pp_group,
+                              get_tensor_model_parallel_world_size)
 from vllm.model_executor.layers.activation import SiluAndMul
 from vllm.model_executor.layers.fused_moe import FusedMoE
 from vllm.model_executor.layers.layernorm import RMSNorm
@@ -51,7 +54,7 @@ from vllm.model_executor.model_loader.weight_utils import (
 from vllm.model_executor.sampling_metadata import SamplingMetadata
 from vllm.sequence import IntermediateTensors
 
-from .interfaces import SupportsPP
+from .interfaces import MixtureOfExperts, SupportsPP
 from .utils import (PPMissingLayer, is_pp_missing_parameter,
                     make_empty_intermediate_tensors_factory, make_layers,
                     maybe_prefix)
@@ -99,11 +102,17 @@ class DeepseekV2MoE(nn.Module):
         config: PretrainedConfig,
         quant_config: Optional[QuantizationConfig] = None,
         prefix: str = "",
+        enable_eplb: bool = False,
     ):
         super().__init__()
         self.tp_size = get_tensor_model_parallel_world_size()
         self.routed_scaling_factor = config.routed_scaling_factor
-        self.n_shared_experts = config.n_shared_experts
+
+        self.ep_group = get_ep_group().device_group
+        self.ep_rank = self.ep_group.rank()
+        self.ep_size = self.ep_group.size()
+        self.n_routed_experts: int = config.n_routed_experts
+        self.n_shared_experts: int = config.n_shared_experts
 
         if config.hidden_act != "silu":
             raise ValueError(f"Unsupported activation: {config.hidden_act}. "
@@ -120,6 +129,22 @@ class DeepseekV2MoE(nn.Module):
         else:
             self.gate.e_score_correction_bias = None
 
+        # Load balancing settings.
+        vllm_config = get_current_vllm_config()
+        parallel_config = vllm_config.parallel_config
+        self.enable_eplb = enable_eplb
+
+        self.n_redundant_experts = parallel_config.num_redundant_experts
+        self.n_logical_experts = self.n_routed_experts
+        self.n_physical_experts = (self.n_logical_experts +
+                                   self.n_redundant_experts)
+        self.n_local_physical_experts = self.n_physical_experts // self.ep_size
+
+        self.physical_expert_start = (self.ep_rank *
+                                      self.n_local_physical_experts)
+        self.physical_expert_end = (self.physical_expert_start +
+                                    self.n_local_physical_experts)
+
         self.experts = FusedMoE(
             num_experts=config.n_routed_experts,
             top_k=config.num_experts_per_tok,
@@ -133,7 +158,9 @@ class DeepseekV2MoE(nn.Module):
             topk_group=config.topk_group,
             prefix=f"{prefix}.experts",
             scoring_func=config.scoring_func,
-            e_score_correction_bias=self.gate.e_score_correction_bias)
+            e_score_correction_bias=self.gate.e_score_correction_bias,
+            enable_eplb=self.enable_eplb,
+            num_redundant_experts=self.n_redundant_experts)
 
         if config.n_shared_experts is not None:
             intermediate_size = (config.moe_intermediate_size *
@@ -503,6 +530,7 @@ class DeepseekV2DecoderLayer(nn.Module):
         model_config: ModelConfig,
         cache_config: Optional[CacheConfig] = None,
         quant_config: Optional[QuantizationConfig] = None,
+        enable_eplb: bool = False,
     ) -> None:
         super().__init__()
         self.hidden_size = config.hidden_size
@@ -543,6 +571,7 @@ class DeepseekV2DecoderLayer(nn.Module):
                 config=config,
                 quant_config=quant_config,
                 prefix=f"{prefix}.mlp",
+                enable_eplb=enable_eplb,
             )
         else:
             self.mlp = DeepseekV2MLP(
@@ -615,6 +644,7 @@ class DeepseekV2Model(nn.Module):
         model_config = vllm_config.model_config
         cache_config = vllm_config.cache_config
         quant_config = vllm_config.quant_config
+        enable_eplb = vllm_config.parallel_config.enable_eplb
         self.config = config
 
         self.vocab_size = config.vocab_size
@@ -636,6 +666,7 @@ class DeepseekV2Model(nn.Module):
                 model_config=model_config,
                 cache_config=cache_config,
                 quant_config=quant_config,
+                enable_eplb=enable_eplb,
             ),
             prefix=f"{prefix}.layers")
 
@@ -681,7 +712,7 @@ class DeepseekV2Model(nn.Module):
         return hidden_states
 
 
-class DeepseekV2ForCausalLM(nn.Module, SupportsPP):
+class DeepseekV2ForCausalLM(nn.Module, SupportsPP, MixtureOfExperts):
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
@@ -700,6 +731,44 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP):
         self.logits_processor = LogitsProcessor(config.vocab_size)
         self.make_empty_intermediate_tensors = (
             self.model.make_empty_intermediate_tensors)
+        self.expert_weights = []
+
+        # Set MoE hyperparameters
+        self.num_moe_layers = (config.num_hidden_layers -
+                               config.first_k_dense_replace)
+        self.num_expert_groups = config.n_group
+
+        self.moe_layers: list[FusedMoE] = []
+        for layer in self.model.layers:
+            assert isinstance(layer, DeepseekV2DecoderLayer)
+            if isinstance(layer.mlp, DeepseekV2MoE):
+                self.moe_layers.append(layer.mlp.experts)
+
+        # Pick last one layer since the first ones may be dense layers.
+        example_moe = typing.cast(
+            DeepseekV2MoE, self.model.layers[config.num_hidden_layers - 1].mlp)
+        self.num_logical_experts = example_moe.n_logical_experts
+        self.num_physical_experts = example_moe.n_physical_experts
+        self.num_local_physical_experts = example_moe.n_local_physical_experts
+        self.num_routed_experts = example_moe.n_routed_experts
+        self.num_shared_experts = example_moe.n_shared_experts
+        self.num_redundant_experts = example_moe.n_redundant_experts
+
+    def set_eplb_state(
+        self,
+        expert_load_view: torch.Tensor,
+        logical_to_physical_map: torch.Tensor,
+        logical_replica_count: torch.Tensor,
+    ) -> None:
+        for layer_idx, layer in enumerate(self.moe_layers):
+            # Register the expert weights.
+            self.expert_weights.append(layer.get_expert_weights())
+            layer.set_eplb_state(
+                moe_layer_idx=layer_idx,
+                expert_load_view=expert_load_view,
+                logical_to_physical_map=logical_to_physical_map,
+                logical_replica_count=logical_replica_count,
+            )
 
     def get_input_embeddings(self, input_ids: torch.Tensor) -> torch.Tensor:
         return self.model.get_input_embeddings(input_ids)
@@ -752,7 +821,8 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP):
             ckpt_gate_proj_name="gate_proj",
             ckpt_down_proj_name="down_proj",
             ckpt_up_proj_name="up_proj",
-            num_experts=self.config.n_routed_experts)
+            num_experts=self.config.n_routed_experts,
+            num_redundant_experts=self.num_redundant_experts)
 
         params_dict = dict(self.named_parameters())
         loaded_params: set[str] = set()
@@ -789,24 +859,44 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP):
                 weight_loader(param, loaded_weight, shard_id)
                 break
             else:
+                is_expert_weight = False
                 for mapping in expert_params_mapping:
                     param_name, weight_name, expert_id, shard_id = mapping
                     if weight_name not in name:
                         continue
-                    name = name.replace(weight_name, param_name)
 
-                    if is_pp_missing_parameter(name, self):
+                    # Anyway, this is an expert weight and should not be
+                    # attempted to load as other weights later
+                    is_expert_weight = True
+
+                    # Do not modify `name` since the loop may continue here
+                    # Instead, create a new variable
+                    name_mapped = name.replace(weight_name, param_name)
+
+                    if is_pp_missing_parameter(name_mapped, self):
                         continue
 
-                    param = params_dict[name]
-                    weight_loader = param.weight_loader
-                    weight_loader(param,
-                                  loaded_weight,
-                                  name,
-                                  shard_id=shard_id,
-                                  expert_id=expert_id)
-                    break
+                    param = params_dict[name_mapped]
+                    # We should ask the weight loader to return success or not
+                    # here since otherwise we may skip experts with other
+                    # available replicas.
+                    weight_loader = typing.cast(Callable[..., bool],
+                                                param.weight_loader)
+                    success = weight_loader(param,
+                                            loaded_weight,
+                                            name_mapped,
+                                            shard_id=shard_id,
+                                            expert_id=expert_id,
+                                            return_success=True)
+                    if success:
+                        break
                 else:
+                    if is_expert_weight:
+                        # We've checked that this is an expert weight
+                        # However it's not mapped locally to this rank
+                        # So we simply skip it
+                        continue
+
                     # Skip loading extra bias for GPTQ models.
                     if name.endswith(".bias") and name not in params_dict:
                         continue
@@ -824,6 +914,7 @@ class DeepseekV2ForCausalLM(nn.Module, SupportsPP):
                                             default_weight_loader)
                     weight_loader(param, loaded_weight)
             loaded_params.add(name)
+
         return loaded_params
 
 

vllm/model_executor/models/interfaces.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
+from collections.abc import Iterable, MutableSequence
 from typing import (TYPE_CHECKING, ClassVar, Literal, Optional, Protocol,
                     Union, overload, runtime_checkable)
 
@@ -426,6 +427,73 @@ def is_hybrid(
     return isinstance(model, IsHybrid)
 
 
+@runtime_checkable
+class MixtureOfExperts(Protocol):
+    """
+    Check if the model is a mixture of experts (MoE) model.
+    """
+
+    expert_weights: MutableSequence[Iterable[Tensor]]
+    """
+    Expert weights saved in this rank.
+
+    The first dimension is the layer, and the second dimension is different
+    parameters in the layer, e.g. up/down projection weights.
+    """
+
+    num_moe_layers: int
+    """Number of MoE layers in this model."""
+
+    num_expert_groups: int
+    """Number of expert groups in this model."""
+
+    num_logical_experts: int
+    """Number of logical experts in this model."""
+
+    num_physical_experts: int
+    """Number of physical experts in this model."""
+
+    num_local_physical_experts: int
+    """Number of local physical experts in this model."""
+
+    num_routed_experts: int
+    """Number of routed experts in this model."""
+
+    num_shared_experts: int
+    """Number of shared experts in this model."""
+
+    num_redundant_experts: int
+    """Number of redundant experts in this model."""
+
+    def set_eplb_state(
+        self,
+        expert_load_view: Tensor,
+        logical_to_physical_map: Tensor,
+        logical_replica_count: Tensor,
+    ) -> None:
+        """
+        Register the EPLB state in the MoE model.
+        
+        Since these are views of the actual EPLB state, any changes made by
+        the EPLB algorithm are automatically reflected in the model's behavior
+        without requiring additional method calls to set new states.
+
+        You should also collect model's `expert_weights` here instead of in
+        the weight loader, since after initial weight loading, further
+        processing like quantization may be applied to the weights.
+
+        Args:
+            expert_load_view: A view of the expert load metrics tensor.
+            logical_to_physical_map: Mapping from logical to physical experts.
+            logical_replica_count: Count of replicas for each logical expert.
+        """
+        ...
+
+
+def is_mixture_of_experts(model: object) -> TypeIs[MixtureOfExperts]:
+    return isinstance(model, MixtureOfExperts)
+
+
 @runtime_checkable
 class HasNoOps(Protocol):
     has_noops: ClassVar[Literal[True]] = True

vllm/v1/worker/gpu_model_runner.py

@@ -21,6 +21,7 @@ from vllm.attention.layer import Attention
 from vllm.compilation.counter import compilation_counter
 from vllm.config import (CompilationLevel, VllmConfig,
                          get_layers_from_vllm_config)
+from vllm.distributed.eplb.eplb_state import EplbState
 from vllm.distributed.kv_transfer import (get_kv_transfer_group,
                                           has_kv_transfer_group)
 from vllm.distributed.kv_transfer.kv_connector.v1 import KVConnectorBase_V1
@@ -33,7 +34,8 @@ from vllm.logger import init_logger
 from vllm.model_executor.layers.mamba.mamba_mixer2 import MambaMixer2
 from vllm.model_executor.layers.rotary_embedding import MRotaryEmbedding
 from vllm.model_executor.model_loader import TensorizerLoader, get_model_loader
-from vllm.model_executor.models.interfaces import has_step_pooler
+from vllm.model_executor.models.interfaces import (has_step_pooler,
+                                                   is_mixture_of_experts)
 from vllm.multimodal import MULTIMODAL_REGISTRY
 from vllm.multimodal.inputs import MultiModalKwargs, PlaceholderRange
 from vllm.multimodal.utils import group_mm_inputs_by_modality
@@ -150,6 +152,13 @@ class GPUModelRunner(LoRAModelRunnerMixin):
         # Sampler
         self.sampler = Sampler()
 
+        self.eplb_state: Optional[EplbState] = None
+        """
+        State of the expert parallelism load balancer.
+
+        Will be lazily initialized when the model is loaded.
+        """
+
         # Lazy initializations
         # self.model: nn.Module  # Set after load_model
         # Initialize in initialize_kv_cache
@@ -1178,6 +1187,24 @@ class GPUModelRunner(LoRAModelRunnerMixin):
             for k, v in self.intermediate_tensors.items()
         })
 
+    def eplb_step(self,
+                  is_dummy: bool = False,
+                  is_profile: bool = False) -> None:
+        """
+        Step for the EPLB (Expert Parallelism Load Balancing) state.
+        """
+        if not self.parallel_config.enable_eplb:
+            return
+
+        assert self.eplb_state is not None
+        assert is_mixture_of_experts(self.model)
+        self.eplb_state.step(
+            self.model,
+            is_dummy,
+            is_profile,
+            log_stats=self.parallel_config.eplb_log_balancedness,
+        )
+
     def get_dp_padding(self,
                        num_tokens: int) -> tuple[int, Optional[torch.Tensor]]:
         dp_size = self.vllm_config.parallel_config.data_parallel_size
@@ -1595,6 +1622,8 @@ class GPUModelRunner(LoRAModelRunnerMixin):
         if has_kv_transfer_group():
             get_kv_transfer_group().clear_connector_metadata()
 
+        self.eplb_step()
+
         return ModelRunnerOutput(
             req_ids=self.input_batch.req_ids,
             req_id_to_index=self.input_batch.req_id_to_index,
@@ -1729,6 +1758,16 @@ class GPUModelRunner(LoRAModelRunnerMixin):
                     time_after_load - time_before_load)
         prepare_communication_buffer_for_model(self.model)
 
+        if is_mixture_of_experts(
+                self.model) and self.parallel_config.enable_eplb:
+            logger.info("EPLB is enabled for model %s.",
+                        self.model_config.model)
+            self.eplb_state = EplbState.build(
+                self.model,
+                self.device,
+                self.parallel_config,
+            )
+
     def save_tensorized_model(
         self,
         tensorizer_config: "TensorizerConfig",
@@ -1887,6 +1926,8 @@ class GPUModelRunner(LoRAModelRunnerMixin):
         self,
         num_tokens: int,
         capture_attn_cudagraph: bool = False,
+        skip_eplb: bool = False,
+        is_profile: bool = False,
     ) -> tuple[torch.Tensor, torch.Tensor]:
 
         # Padding for DP
@@ -1983,6 +2024,16 @@ class GPUModelRunner(LoRAModelRunnerMixin):
                 assert isinstance(self.drafter, EagleProposer)
                 self.drafter.dummy_run(num_tokens)
 
+        # This is necessary to avoid blocking DP.
+        # For dummy runs, we typically skip EPLB since we don't have any real
+        # requests to process.
+        # However, in DP settings, there may be cases when some DP ranks do
+        # not have any requests to process, so they're executing dummy batches.
+        # In such cases, we still have to trigger EPLB to make sure
+        # ranks execute the rearrangement in synchronization.
+        if not skip_eplb:
+            self.eplb_step(is_dummy=True, is_profile=is_profile)
+
         logit_indices = np.cumsum(num_scheduled_tokens) - 1
         return hidden_states, hidden_states[logit_indices]
 
@@ -2175,8 +2226,9 @@ class GPUModelRunner(LoRAModelRunnerMixin):
             # Cache the dummy encoder outputs.
             self.encoder_cache["tmp"] = dict(enumerate(dummy_encoder_outputs))
 
+        # Add `is_profile` here to pre-allocate communication buffers
         hidden_states, last_hidden_states \
-            = self._dummy_run(self.max_num_tokens)
+            = self._dummy_run(self.max_num_tokens, is_profile=True)
         if get_pp_group().is_last_rank:
             if self.is_pooling_model:
                 output = self._dummy_pooler_run(hidden_states)
@@ -2210,10 +2262,15 @@ class GPUModelRunner(LoRAModelRunnerMixin):
             for num_tokens in tqdm(reversed(self.cudagraph_batch_sizes),
                                    desc="Capturing CUDA graphs",
                                    total=len(self.cudagraph_batch_sizes)):
+                # We skip EPLB here since we don't want to record dummy metrics
                 for _ in range(
                         self.compilation_config.cudagraph_num_of_warmups):
-                    self._dummy_run(num_tokens, capture_attn_cudagraph=full_cg)
-                self._dummy_run(num_tokens, capture_attn_cudagraph=full_cg)
+                    self._dummy_run(num_tokens,
+                                    capture_attn_cudagraph=full_cg,
+                                    skip_eplb=True)
+                self._dummy_run(num_tokens,
+                                capture_attn_cudagraph=full_cg,
+                                skip_eplb=True)
 
         end_time = time.perf_counter()
         end_free_gpu_memory = torch.cuda.mem_get_info()[0]

vllm/v1/worker/gpu_worker.py

@@ -259,9 +259,10 @@ class Worker(WorkerBase):
                 x for x in warmup_sizes if x not in
                 self.vllm_config.compilation_config.cudagraph_capture_sizes
             ]
+        # We skip EPLB here since we don't want to record dummy metrics
         for size in sorted(warmup_sizes, reverse=True):
             logger.info("Compile and warming up model for size %d", size)
-            self.model_runner._dummy_run(size)
+            self.model_runner._dummy_run(size, skip_eplb=True)
         if not self.model_config.enforce_eager:
             self.model_runner.capture_model()
 
@@ -274,8 +275,12 @@ class Worker(WorkerBase):
             max_num_reqs = min(self.scheduler_config.max_num_seqs,
                                self.scheduler_config.max_num_batched_tokens)
 
+            # We skip EPLB here since we don't want to record dummy metrics
             hidden_states, last_hidden_states = \
-                self.model_runner._dummy_run(num_tokens=max_num_reqs)
+                self.model_runner._dummy_run(
+                    num_tokens=max_num_reqs,
+                    skip_eplb=True,
+                )
             if self.model_runner.is_pooling_model:
                 self.model_runner._dummy_pooler_run(hidden_states)
             else: