[Misc] Add override for allreduce fusion thresholds

vllm/compilation/collective_fusion.py

@@ -10,6 +10,7 @@
 from torch._inductor.pattern_matcher import PatternMatcherPass
 from torch.distributed._symmetric_memory import enable_symm_mem_for_group
 
+import vllm.envs as envs
 from vllm.config import VllmConfig
 from vllm.distributed import get_tp_group, tensor_model_parallel_all_reduce
 from vllm.distributed.parallel_state import (
@@ -401,6 +402,18 @@ def __call__(self, graph: fx.Graph):
         6: MiB // 2,  # 512KB
         8: MiB // 2,  # 512KB
     }
+
+    try:
+        _FI_MAX_SIZES.update({
+            int(k): int(float(v) * MiB)
+            for k, v in
+            envs.VLLM_FLASHINFER_ALLREDUCE_FUSION_THRESHOLDS_MB.items()
+        })
+    except Exception as e:
+        raise ValueError(
+            "Failed to parse VLLM_FLASHINFER_ALLREDUCE_FUSION_THRESHOLDS_MB: "
+            + str(e)) from e
+
     # opt for a more conservative default value
     # when world size is not in _FI_MAX_SIZES
     _DEFAULT_FI_MAX_SIZE = MiB // 2

vllm/envs.py

@@ -2,6 +2,7 @@
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 
 import hashlib
+import json
 import os
 import sys
 import tempfile
@@ -1046,6 +1047,16 @@ def get_vllm_port() -> Optional[int]:
     "VLLM_MAX_TOKENS_PER_EXPERT_FP4_MOE":
     lambda: int(os.getenv("VLLM_MAX_TOKENS_PER_EXPERT_FP4_MOE", "163840")),
 
+    # Specifies the thresholds of the communicated tensor sizes under which
+    # vllm should use flashinfer fused allreduce. The variable should be a
+    # JSON with the following format:
+    #     { <world size>: <max size in mb> }
+    # Unspecified world sizes will fallback to
+    #     { 2: 64, 4: 1, <everything else>: 0.5 }
+    "VLLM_FLASHINFER_ALLREDUCE_FUSION_THRESHOLDS_MB":
+    lambda: json.loads(os.getenv(
+        "VLLM_FLASHINFER_ALLREDUCE_FUSION_THRESHOLDS_MB", "{}")),
+
     # MoE routing strategy selector.
     # See `RoutingSimulator.get_available_strategies()` # for available
     # strategies.