mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
f430a4b447
* Update TensorRT-LLM --------- Co-authored-by: IbrahimAmin <ibrahimamin532@gmail.com> Co-authored-by: Fabian Joswig <fjosw@users.noreply.github.com> Co-authored-by: Pzzzzz <hello-cd.plus@hotmail.com> Co-authored-by: CoderHam <hemant@cohere.com> Co-authored-by: Konstantin Lopuhin <kostia.lopuhin@gmail.com>
63 lines
1.9 KiB
C++
63 lines
1.9 KiB
C++
/*
|
|
* Copyright (c) 2020-2024, 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.
|
|
*/
|
|
#pragma once
|
|
|
|
#include <cuda_fp16.h>
|
|
|
|
#include "tensorrt_llm/common/cudaUtils.h"
|
|
#include "tensorrt_llm/kernels/penaltyTypes.h"
|
|
#include "tensorrt_llm/runtime/common.h"
|
|
|
|
namespace tensorrt_llm
|
|
{
|
|
namespace kernels
|
|
{
|
|
|
|
template <typename T>
|
|
struct InvokeBatchApplyPenaltyParams
|
|
{
|
|
T const* const* inputLogits;
|
|
T* outputLogits;
|
|
T const* biases;
|
|
runtime::TokenIdType* penaltyWorkspace;
|
|
runtime::TokenIdType const* penaltyWorkspacePrev;
|
|
float const* temperatures;
|
|
float const* repetitionPenalties;
|
|
float const* presencePenalties;
|
|
float const* frequencyPenalties;
|
|
runtime::SizeType32 batchSize;
|
|
runtime::SizeType32 beamWidth;
|
|
runtime::SizeType32 maxSeqLen;
|
|
runtime::SizeType32 vocabSize;
|
|
runtime::SizeType32 vocabSizePadded;
|
|
runtime::TokenIdType const** outputIdsPtr;
|
|
runtime::SizeType32 const** parentIdsPtr;
|
|
runtime::SizeType32 const* inputLengths;
|
|
runtime::SizeType32 const* sequenceLengths;
|
|
runtime::SizeType32 const* minLengths;
|
|
runtime::TokenIdType const* endIds;
|
|
runtime::SizeType32 const* batchSlots;
|
|
runtime::SizeType32 maxTokensPerStep;
|
|
runtime::SizeType32 const* tokensPerStep;
|
|
cudaStream_t stream;
|
|
};
|
|
|
|
template <typename T>
|
|
void invokeBatchApplyPenalty(InvokeBatchApplyPenaltyParams<T> const& params);
|
|
|
|
} // namespace kernels
|
|
} // namespace tensorrt_llm
|