EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Refactor the ops PyTorch adapter，cleanup for csrc/torch_binding.cpp (#6732)

Browse Source 
### What this PR does / why we need it?
Refactor the ops PyTorch adapter，cleanup for csrc/torch_binding.cpp,
more details see
https://github.com/vllm-project/vllm-ascend/issues/6486

### Does this PR introduce _any_ user-facing change?
No

### How was this patch tested?
install the new package to test the new modification, here is the
result:


- vLLM version: v0.15.0
- vLLM main:
9562912cea

---------

Signed-off-by: liziyu <liziyu16@huawei.com>
Signed-off-by: wangxiaoteng <wangxiaoteng@huawei.com>
Signed-off-by: luomin2005 <luomin2005@huawei.com>
Co-authored-by: liziyu <56102866+liziyu179@users.noreply.github.com>
Co-authored-by: wangxiaoteng <wangxiaoteng@huawei.com>
This commit is contained in:
luomin2005
2026-02-24 09:12:43 +08:00
committed by GitHub
parent f0caeeadcb
commit f41eeeb11e
15 changed files with 1037 additions and 735 deletions
Download Patch File Download Diff File Expand all files Collapse all files

72
csrc/moe_combine_normal/moe_combine_normal_torch_adpt.h Normal file
View File

@@ -0,0 +1,72 @@
/*
* Copyright (c) Huawei Technologies Co., Ltd. 2026. All rights reserved.
*
* 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.
*/
#ifndef MOE_COMBINE_NORMAL_TORCH_ADPT_H
#define MOE_COMBINE_NORMAL_TORCH_ADPT_H
namespace vllm_ascend {
at::Tensor combine_prefill(const at::Tensor& x, const at::Tensor& topk_idx, const at::Tensor& topk_weights,
const at::Tensor& src_idx, const at::Tensor& send_head, c10::string_view groupEp,
int64_t rank, int64_t num_ranks) {
std::vector<char> group_ep_chrs(groupEp.begin(), groupEp.end());
group_ep_chrs.push_back('\0');
char* group_ep_ptr = &group_ep_chrs[0];
TORCH_BIND_ASSERT(x.dim() == 2 and x.is_contiguous());
at::Tensor recv_x = x;
at::Tensor topk_idx_p = topk_idx;
auto topk_idx_int32 = topk_idx_p.to(at::kInt);
at::Tensor expand_ids = topk_idx_int32;
at::Tensor token_src_info = src_idx;
at::Tensor ep_send_counts = send_head;
auto device = x.device();
const int num_tokens = topk_idx_p.size(0);
const int num_topk = topk_idx_p.size(1);
int64_t hidden = static_cast<int>(recv_x.size(1));
at::Tensor tp_send_counts = at::empty({1}, at::dtype(at::kInt).device(device));
int64_t tp_world_size = 1;
int64_t tp_rankId = 0;
int64_t moe_expert_number = send_head.size(0);
int64_t global_bs = topk_idx_p.size(0) * num_ranks;
// Combine data
auto combined_x = torch::empty({topk_weights.size(0), hidden}, x.options());
EXEC_NPU_CMD(aclnnMoeCombineNormal,
recv_x,
token_src_info,
ep_send_counts,
topk_weights,
tp_send_counts,
group_ep_ptr,
num_ranks,
rank,
group_ep_ptr,
tp_world_size,
tp_rankId,
moe_expert_number,
global_bs,
combined_x);
return combined_x;
}
}
#endif