[Misc] Add override for allreduce fusion thresholds (#23639)
Signed-off-by: Julien Lin <jullin@nvidia.com>
This commit is contained in:
@ -10,6 +10,7 @@ from torch._higher_order_ops.auto_functionalize import auto_functionalized
|
||||
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 @@ if flashinfer_comm is not None:
|
||||
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
|
||||
|
11
vllm/envs.py
11
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 @@ environment_variables: dict[str, Callable[[], Any]] = {
|
||||
"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.
|
||||
|
Reference in New Issue
Block a user