c59d69d9e6
### What this PR does / why we need it? This PR introduces an expert rearrange algorithm for PanguProMoE model. Different from the original grouped topk, it filters out the top experts that are allocated more tokens. Therefore, we can load less experts when calculating gmm. We have test this algorithm for PanguProMoE-72B on 300I Duo platform and 800I A2 platform. On 300I Duo platform, we find that `num_voted_experts` set to 5 achieves both good performance and accuracy. While on 800I A2, we still set it to 8 to use original pangu grouped topk. ### Does this PR introduce _any_ user-facing change? No. ### How was this patch tested? <!-- CI passed with new added/existing test. If it was tested in a way different from regular unit tests, please clarify how you tested step by step, ideally copy and paste-able, so that other reviewers can test and check, and descendants can verify in the future. If tests were not added, please describe why they were not added and/or why it was difficult to add. --> Signed-off-by: angazenn <zengyanjia@huawei.com> Co-authored-by: angazenn <zengyanjia@huawei.com>
86 lines
2.8 KiB
Python
86 lines
2.8 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.model_executor.layers.fused_moe.layer import \
|
|
UnquantizedFusedMoEMethod
|
|
|
|
from vllm_ascend.ops.fused_moe import (fused_experts, fused_experts_moge,
|
|
select_experts)
|
|
from vllm_ascend.utils import is_310p
|
|
|
|
|
|
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",
|
|
) -> 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,
|
|
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)
|
|
|
|
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)
|
|
|
|
|
|
UnquantizedFusedMoEMethod.forward_oot = forward_oot
|