TensorRT-LLMs/cpp/tensorrt_llm/thop/fusedTopkSoftmax.cpp
Yihan Wang 9df4dad3b6
[None][fix] Introduce inline namespace to avoid symbol collision (#9541)
Signed-off-by: Yihan Wang <yihwang@nvidia.com>
2025-12-12 23:32:15 +08:00

75 lines
2.7 KiB
C++

/*
* Copyright (c) 2020-2023, NVIDIA CORPORATION. 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.
*/
#include "tensorrt_llm/common/cudaUtils.h"
#include "tensorrt_llm/common/workspace.h"
#include "tensorrt_llm/runtime/torchUtils.h"
#include "tensorrt_llm/thop/thUtils.h"
#include <ATen/cuda/EmptyTensor.h>
#include <cuda_fp16.h>
#include <cstdint>
TRTLLM_NAMESPACE_BEGIN
namespace torch_ext
{
std::tuple<torch::Tensor, torch::Tensor> fused_topk_softmax(torch::Tensor const& router_logits, int64_t const top_k,
int64_t const num_experts_total, int64_t const start_expert, int64_t const end_expert)
{
// TODO: enable once the kernel has been added to the internal CUTLASS library.
TLLM_CHECK_WITH_INFO(false, "Fused topk/softmax op has not been enabled yet.");
CHECK_INPUT(router_logits, torch::kBFloat16);
auto const& router_logits_shape = router_logits.sizes();
auto const& rank = router_logits_shape.size();
TORCH_CHECK(rank == 2, "router_logits should be 2D tensor.");
int64_t const num_rows = router_logits_shape[0];
auto token_final_scales
= torch::empty({num_rows, top_k}, torch::dtype(torch::kFloat32).device(router_logits.device()));
auto token_selected_experts
= torch::empty({num_rows, top_k}, torch::dtype(torch::kInt32).device(router_logits.device()));
// auto stream = at::cuda::getCurrentCUDAStream(router_logits.get_device());
// tensorrt_llm::kernels::topkGatingSoftmaxKernelLauncher(
// static_cast<__nv_bfloat16 const*>(router_logits.const_data_ptr()),
// static_cast<float*>(token_final_scales.data_ptr()), static_cast<int*>(token_selected_experts.data_ptr()),
// num_rows, top_k, num_experts_total, start_expert, end_expert, stream);
return {token_final_scales, token_selected_experts};
}
} // namespace torch_ext
TRTLLM_NAMESPACE_END
TORCH_LIBRARY_FRAGMENT(trtllm, m)
{
m.def(
"fused_topk_softmax(Tensor router_logits, int top_k, "
"int num_experts_total, int start_expert, "
"int end_expert) -> (Tensor, Tensor) ");
}
TORCH_LIBRARY_IMPL(trtllm, CUDA, m)
{
m.impl("fused_topk_softmax", &tensorrt_llm::torch_ext::fused_topk_softmax);
}