a9480d5f0a
### What this PR does / why we need it?
Updates the FusedMoE method to determine whether to use ACL Graph based
on the `torchair_graph_config`
This is equivalent to #2154 on v0.9.1-dev.
### Does this PR introduce _any_ user-facing change?
None.
### How was this patch tested?
None needed.
- vLLM version: v0.10.0
- vLLM main:
ad57f23f6a
Signed-off-by: Yizhou Liu <liu_yizhou@outlook.com>
116 lines
4.2 KiB
Python
116 lines
4.2 KiB
Python
#
|
|
# Copyright (c) 2025 Huawei Technologies Co., Ltd. All Rights Reserved.
|
|
# This file is a part of the vllm-ascend project.
|
|
#
|
|
# Licensed under the Apache License, Version 2.0 (the "License");
|
|
# you may not use this file except in compliance with the License.
|
|
# You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing, software
|
|
# distributed under the License is distributed on an "AS IS" BASIS,
|
|
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
# See the License for the specific language governing permissions and
|
|
# limitations under the License.
|
|
#
|
|
|
|
from typing import Callable, Optional
|
|
|
|
import torch
|
|
from vllm.config import CompilationLevel, get_current_vllm_config
|
|
from vllm.model_executor.layers.fused_moe.layer import \
|
|
UnquantizedFusedMoEMethod
|
|
|
|
from vllm_ascend.ascend_config import get_ascend_config
|
|
from vllm_ascend.ops.fused_moe import (fused_experts, fused_experts_moge,
|
|
select_experts)
|
|
from vllm_ascend.utils import is_310p
|
|
|
|
original_unquantized_fused_moe_init_func = UnquantizedFusedMoEMethod.__init__
|
|
|
|
|
|
def unquantized_fused_moe_init_func(self, *args, **kwargs):
|
|
original_unquantized_fused_moe_init_func(self, *args, **kwargs)
|
|
vllm_config = get_current_vllm_config()
|
|
self.max_num_batched_tokens = vllm_config.scheduler_config.max_num_batched_tokens
|
|
|
|
ascend_config = get_ascend_config()
|
|
|
|
if ascend_config.torchair_graph_config.enabled:
|
|
self.use_aclgraph = False
|
|
else:
|
|
self.use_aclgraph = (vllm_config.compilation_config.level
|
|
== CompilationLevel.PIECEWISE
|
|
and not vllm_config.model_config.enforce_eager)
|
|
|
|
|
|
def forward_oot(
|
|
self,
|
|
layer: torch.nn.Module,
|
|
x: torch.Tensor,
|
|
use_grouped_topk: bool,
|
|
top_k: int,
|
|
router_logits: torch.Tensor,
|
|
renormalize: bool,
|
|
topk_group: Optional[int] = None,
|
|
num_expert_group: Optional[int] = None,
|
|
custom_routing_function: Optional[Callable] = None,
|
|
scoring_func: str = "softmax",
|
|
e_score_correction_bias: Optional[torch.Tensor] = None,
|
|
global_num_experts: Optional[int] = None,
|
|
expert_map: Optional[torch.Tensor] = None,
|
|
apply_router_weight_on_input: bool = False,
|
|
activation: str = "silu",
|
|
enable_eplb: bool = False,
|
|
expert_load_view: Optional[torch.Tensor] = None,
|
|
logical_to_physical_map: Optional[torch.Tensor] = None,
|
|
logical_replica_count: Optional[torch.Tensor] = None) -> torch.Tensor:
|
|
|
|
topk_weights, topk_ids = select_experts(
|
|
global_num_experts=global_num_experts,
|
|
hidden_states=x,
|
|
router_logits=router_logits,
|
|
top_k=top_k,
|
|
use_grouped_topk=use_grouped_topk,
|
|
renormalize=renormalize,
|
|
topk_group=topk_group,
|
|
num_expert_group=num_expert_group,
|
|
custom_routing_function=custom_routing_function,
|
|
scoring_func=scoring_func,
|
|
e_score_correction_bias=e_score_correction_bias,
|
|
)
|
|
|
|
if topk_ids.shape[1] < top_k or is_310p():
|
|
assert global_num_experts is not None
|
|
return fused_experts_moge(
|
|
hidden_states=x,
|
|
w1=layer.w13_weight,
|
|
w2=layer.w2_weight,
|
|
moe_parallel_config=self.moe.moe_parallel_config,
|
|
topk_weights=topk_weights,
|
|
topk_ids=topk_ids,
|
|
top_k=top_k,
|
|
global_num_experts=global_num_experts,
|
|
expert_map=expert_map,
|
|
apply_router_weight_on_input=apply_router_weight_on_input)
|
|
|
|
# If use aclgraph, we need to set max_num_tokens to make
|
|
# the input shape of `npu_moe_init_routing` fixed
|
|
max_num_tokens = self.max_num_batched_tokens if self.use_aclgraph else None
|
|
|
|
return fused_experts(
|
|
hidden_states=x,
|
|
w1=layer.w13_weight,
|
|
w2=layer.w2_weight,
|
|
topk_weights=topk_weights,
|
|
topk_ids=topk_ids,
|
|
top_k=top_k,
|
|
expert_map=expert_map,
|
|
apply_router_weight_on_input=apply_router_weight_on_input,
|
|
max_num_tokens=max_num_tokens)
|
|
|
|
|
|
UnquantizedFusedMoEMethod.__init__ = unquantized_fused_moe_init_func
|
|
UnquantizedFusedMoEMethod.forward_oot = forward_oot
|