Fix Flashinfer CUTLASS MOE Allgather (#21963)

Signed-off-by: Shu Wang <shuw@nvidia.com>

vllm/distributed/device_communicators/cuda_communicator.py

@@ -236,7 +236,8 @@ class CudaCommunicator(DeviceCommunicatorBase):
             input_size = input_.size()
             if sizes is not None:
                 assert len(sizes) == world_size
-                assert input_.shape[dim] == sizes[self.rank_in_group]
+                assert input_.shape[dim] == sizes[self.rank_in_group], (
+                    f"{input_.shape[dim]} != {sizes[self.rank_in_group]}")
                 output_size = (sum(sizes), ) + input_size[1:]
             else:
                 output_size = (input_size[0] * world_size, ) + input_size[1:]

vllm/forward_context.py

@@ -26,10 +26,26 @@ batchsize_logging_interval: float = envs.VLLM_LOG_BATCHSIZE_INTERVAL
 batchsize_forward_time: defaultdict = defaultdict(list)
 
 
+def _compute_chunked_local_num_tokens(num_tokens_across_dp_cpu: list[int],
+                                      max_num_tokens: int,
+                                      chunk_idx: int) -> list[int]:
+    dp_size = len(num_tokens_across_dp_cpu)
+
+    local_size = [-1] * dp_size
+    for i in range(dp_size):
+        dp_tokens = num_tokens_across_dp_cpu[i]
+        local_size[i] = min(max_num_tokens,
+                            dp_tokens - (max_num_tokens * chunk_idx))
+        if local_size[i] <= 0:
+            local_size[i] = 1  # ensure lockstep even if done
+    return local_size
+
+
 @dataclass
 class DPMetadata:
     max_tokens_across_dp_cpu: torch.Tensor
     cu_tokens_across_dp_cpu: torch.Tensor
+    local_sizes: Optional[list[int]] = None
 
     @staticmethod
     def num_tokens_across_dp(num_tokens: int, dp_size: int,
@@ -78,6 +94,48 @@ class DPMetadata:
         cu_tokens_across_dp_cpu = torch.cumsum(num_tokens_across_dp, dim=0)
         return DPMetadata(max_tokens_across_dp_cpu, cu_tokens_across_dp_cpu)
 
+    @contextmanager
+    def chunked_sizes(self, max_chunk_size_per_rank: int, chunk_idx: int):
+        """
+        Context manager to compute and temporarily set the per-rank local token
+        sizes for a specific chunk during chunked forward execution.
+
+        This is necessary to ensure each DP (data parallel) rank processes its
+        designated portion of tokens in lockstep with others, even when the
+        token counts are uneven or some ranks have completed their input early.
+
+        For chunked execution, we break up the total tokens on each rank into
+        multiple chunks (of at most `max_chunk_size_per_rank`), and for a given
+        `chunk_idx`, this context manager sets `self.local_sizes` to the number
+        of tokens to process in that chunk on each rank.
+
+        It uses cumulative sizes (`cu_tokens_across_dp_cpu`) to derive the
+        number of tokens per rank, and calls `_compute_chunked_local_num_tokens`
+        to determine the chunk-wise split.
+
+        `self.local_sizes` is only valid inside the context.
+
+        Args:
+            max_chunk_size_per_rank: The max number of tokens each rank is 
+                                     allowed to process in this chunk.
+            chunk_idx: The index of the chunk to compute sizes for.
+        """
+        cu_sizes = self.cu_tokens_across_dp_cpu
+        num_tokens_across_dp_cpu = [
+            (cu_sizes[i] -
+             cu_sizes[i - 1]).item() if i > 0 else cu_sizes[0].item()
+            for i in range(len(cu_sizes))
+        ]
+        self.local_sizes = _compute_chunked_local_num_tokens(
+            num_tokens_across_dp_cpu, max_chunk_size_per_rank, chunk_idx)
+        try:
+            yield self.local_sizes
+        finally:
+            self.local_sizes = None
+
+    def get_chunk_sizes_across_dp_rank(self) -> Optional[list[int]]:
+        return self.local_sizes
+
 
 @dataclass
 class ForwardContext:

vllm/model_executor/layers/fused_moe/flashinfer_cutlass_prepare_finalize.py

@@ -4,7 +4,6 @@ from typing import Any, Optional
 
 import torch
 
-import vllm.envs as envs
 import vllm.model_executor.layers.fused_moe.modular_kernel as mk
 from vllm.distributed import get_dp_group
 from vllm.forward_context import get_forward_context
@@ -14,20 +13,8 @@ from vllm.model_executor.layers.fused_moe.utils import (
 from vllm.utils.flashinfer import nvfp4_block_scale_interleave
 
 
-def get_local_sizes(local_tokens):
-    cu_sizes = get_forward_context().dp_metadata.cu_tokens_across_dp_cpu
-    sizes = [cu_sizes[0].item()]
-    for i in range(1, len(cu_sizes)):
-        sizes.append((cu_sizes[i] - cu_sizes[i - 1]).item())
-    max_num_tokens = envs.VLLM_MOE_DP_CHUNK_SIZE
-    sizes_chunked = [max_num_tokens] * len(sizes)
-    if local_tokens < max_num_tokens:
-        # When the number of local tokens is less than max_num_tokens, all other
-        # ranks will also have fewer than max_num_tokens. The remaining tokens
-        # are accounted for as residual.
-        sizes_chunked = [x % max_num_tokens for x in sizes]
-
-    return sizes_chunked
+def get_local_sizes():
+    return get_forward_context().dp_metadata.get_chunk_sizes_across_dp_rank()
 
 
 class FlashInferCutlassMoEPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):
@@ -90,7 +77,7 @@ class FlashInferCutlassMoEPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):
             topk_weights, topk_ids, a1q, a1q_scale = \
                 get_dp_group().all_gatherv([topk_weights, topk_ids, a1q, a1q_scale], # noqa: E501
                                            dim=0,
-                                           sizes=get_local_sizes(local_tokens))
+                                           sizes=get_local_sizes())
             a1_m, a1_n = a1q.shape
             a1q_scale = nvfp4_block_scale_interleave(a1q_scale)
 
@@ -107,8 +94,5 @@ class FlashInferCutlassMoEPrepareAndFinalize(mk.FusedMoEPrepareAndFinalize):
                                                ['use_dp', 'local_tokens'])
         if use_dp:
             fused_expert_output = get_dp_group().reduce_scatterv(
-                fused_expert_output,
-                dim=0,
-                sizes=get_local_sizes(local_tokens),
-            )
+                fused_expert_output, dim=0, sizes=get_local_sizes())
         output.copy_(fused_expert_output)

vllm/model_executor/layers/fused_moe/layer.py

@@ -1570,18 +1570,19 @@ class FusedMoE(torch.nn.Module):
         max_tokens_across_dp = ctx.dp_metadata.max_tokens_across_dp_cpu
         moe_dp_chunk_size_per_rank = self.moe_config.max_num_tokens
         num_tokens = full_hidden_states.size(0)
-        for chunk_start_ in range(0, max_tokens_across_dp,
-                                  moe_dp_chunk_size_per_rank):
+        for chunk_idx, chunk_start_ in enumerate(
+                range(0, max_tokens_across_dp, moe_dp_chunk_size_per_rank)):
             chunk_start = chunk_start_
             chunk_end = min(chunk_start + moe_dp_chunk_size_per_rank,
                             max_tokens_across_dp)
             # clamp start and end
             chunk_start = min(chunk_start, num_tokens - 1)
             chunk_end = min(chunk_end, num_tokens)
-
-            process_chunk(chunk_start,
-                          chunk_end,
-                          skip_result_store=chunk_start_ >= num_tokens)
+            with ctx.dp_metadata.chunked_sizes(moe_dp_chunk_size_per_rank,
+                                               chunk_idx):
+                process_chunk(chunk_start,
+                              chunk_end,
+                              skip_result_store=chunk_start_ >= num_tokens)
 
         return full_final_hidden_states