TensorRT-LLMs/cpp/tensorrt_llm/layers/beamSearchLayer.h
2024-04-30 17:19:10 +08:00

124 lines
4.1 KiB
C++

/*
* Copyright (c) 2019-2024, 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 "tensorrt_llm/common/tensor.h"
#include "tensorrt_llm/kernels/beamSearchKernels.h"
#include "tensorrt_llm/kernels/decodingCommon.h"
#include "tensorrt_llm/layers/baseLayer.h"
#include "tensorrt_llm/layers/decodingParams.h"
#include "tensorrt_llm/runtime/common.h"
#include <utility>
#include <optional>
namespace tc = tensorrt_llm::common;
namespace tensorrt_llm
{
namespace layers
{
// BS: batch_size, lBS: local_batch_size, BM: beam_width, mSL: max_seq_length
class BeamSearchSetupParams : public BaseSetupParams
{
public:
std::optional<std::vector<float>> beam_search_diversity_rate; // [BS] on cpu
std::optional<std::vector<float>> length_penalty; // [BS] on cpu
std::optional<std::vector<int>> early_stopping; // [BS] on cpu
};
class BeamSearchInputParams : public BaseInputParams
{
public:
explicit BeamSearchInputParams(runtime::SizeType step, runtime::SizeType ite, tc::Tensor logits, tc::Tensor endIds,
tc::Tensor src_cache_indirection, runtime::SizeType max_attention_window, runtime::SizeType sink_token_length,
runtime::SizeType max_seq_len)
: BaseInputParams(step, ite, std::move(endIds))
, logits{std::move(logits)}
, max_attention_window{max_attention_window}
, sink_token_length{sink_token_length}
, max_seq_len{max_seq_len}
, src_cache_indirection{std::move(src_cache_indirection)}
{
}
// mandatory parameters
tc::Tensor logits; // [maxBatchSize, beamWidth, vocabSizePadded]
runtime::SizeType max_attention_window;
runtime::SizeType sink_token_length;
runtime::SizeType max_seq_len;
tc::Tensor src_cache_indirection; // [BS, BM, mSL]
std::optional<tc::Tensor> input_lengths; // [BS, BM]
};
class BeamSearchOutputParams : public BaseOutputParams
{
public:
explicit BeamSearchOutputParams(tc::Tensor outputIds, tc::Tensor parentIds, tc::Tensor tgt_cache_indirection)
: BaseOutputParams{std::move(outputIds)}
, parent_ids{std::move(parentIds)}
, tgt_cache_indirection{std::move(tgt_cache_indirection)}
{
}
std::shared_ptr<kernels::BeamHypotheses> beamHypotheses;
tc::Tensor parent_ids; // [BS, BM, mSL]
tc::Tensor tgt_cache_indirection; // [BS, BM, mSL]
tc::Tensor parent_ids_ptr; // [BS][BM, mSL]
};
template <typename T>
class BeamSearchLayer : public BaseLayer
{
using Base = BaseLayer;
public:
BeamSearchLayer(DecoderDomain const& decoderDomain, cudaStream_t stream, std::shared_ptr<tc::IAllocator> allocator);
~BeamSearchLayer() override;
void setup(runtime::SizeType const batch_size, runtime::SizeType const beamWidth,
runtime::SizeType const* batchSlots, std::shared_ptr<BaseSetupParams> setupParams) override;
void forward(std::shared_ptr<BaseOutputParams> outputs, std::shared_ptr<BaseInputParams> inputs) override;
private:
using Base::mAllocator;
using Base::mStream;
bool mIsAllocateBuffer;
runtime::SizeType mVocabSize{0};
runtime::SizeType mVocabSizePadded{0};
size_t mWorkspaceSize{0};
void* mWorkspace{nullptr};
// TODO: use pinned memory to simplify the buffers?
float* mDiversityRateDevice;
float* mLengthPenaltyDevice;
int* mEarlyStoppingDevice;
std::vector<float> mDiversityRateHost;
std::vector<float> mLengthPenaltyHost;
std::vector<int> mEarlyStoppingHost;
void allocateBuffer(runtime::SizeType const batch_size, runtime::SizeType const beam_width);
void freeBuffer();
};
} // namespace layers
} // namespace tensorrt_llm