mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
2a115dae84
Co-authored-by: DreamGenX <x@dreamgen.com> Co-authored-by: Ace-RR <78812427+Ace-RR@users.noreply.github.com> Co-authored-by: bprus <39293131+bprus@users.noreply.github.com> Co-authored-by: janpetrov <janpetrov@icloud.com>
71 lines
2.7 KiB
C++
71 lines
2.7 KiB
C++
/*
|
|
* Copyright (c) 2019-2024, NVIDIA CORPORATION. All rights reserved.
|
|
* Copyright (c) 2021, NAVER Corp. Authored by CLOVA.
|
|
*
|
|
* 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/executor/types.h"
|
|
#include "tensorrt_llm/layers/baseLayer.h"
|
|
#include "tensorrt_llm/layers/decodingParams.h"
|
|
|
|
#include <curand_kernel.h>
|
|
|
|
namespace tensorrt_llm::layers
|
|
{
|
|
|
|
//! \brief Layer to process stop criteria. Supports:
|
|
//! 1. Stop words criteria
|
|
//! 2. Maximum length criteria
|
|
template <typename T>
|
|
class StopCriteriaLayer : public BaseLayer
|
|
{
|
|
public:
|
|
StopCriteriaLayer(executor::DecodingMode const& mode, DecoderDomain const& /* decoderDomain */, cudaStream_t stream,
|
|
std::shared_ptr<tensorrt_llm::common::IAllocator> allocator);
|
|
|
|
~StopCriteriaLayer() override = default;
|
|
|
|
void setup(runtime::SizeType32 batchSize, runtime::SizeType32 beamWidth, runtime::SizeType32 const* batchSlots,
|
|
std::shared_ptr<BaseSetupParams> const& setupParams) override;
|
|
|
|
void forwardAsync(std::shared_ptr<BaseDecodingOutputs> const& outputs,
|
|
std::shared_ptr<BaseDecodingInputs> const& inputs) override;
|
|
|
|
private:
|
|
static void checkMaxLengthStopCriteria(std::shared_ptr<BaseDecodingOutputs>& outputs,
|
|
std::shared_ptr<DecodingInputs> const& inputs, runtime::SizeType32 const* batchSlots,
|
|
DecoderDomain const& decoderDomain, runtime::SizeType32 maxSeqLen, cudaStream_t stream);
|
|
static void checkStopWordsStopCriteria(std::shared_ptr<BaseDecodingOutputs>& outputs,
|
|
std::shared_ptr<DecodingInputs> const& inputs, runtime::SizeType32 const* batchSlots,
|
|
DecoderDomain const& decoderDomain, runtime::SizeType32 maxSeqLen, cudaStream_t stream);
|
|
static void checkEosToken(std::shared_ptr<BaseDecodingOutputs>& outputs,
|
|
std::shared_ptr<DecodingInputs> const& inputs, runtime::SizeType32 const* batchSlots,
|
|
DecoderDomain const& decoderDomain, runtime::SizeType32 maxSeqLen, cudaStream_t stream);
|
|
|
|
private:
|
|
using BaseLayer::mWorkspaceSize;
|
|
using BaseLayer::mAllocatedSize;
|
|
|
|
using BaseLayer::mStream;
|
|
using BaseLayer::mAllocator;
|
|
|
|
using BaseLayer::mDecoderDomain;
|
|
|
|
executor::DecodingMode mDecodingMode;
|
|
};
|
|
|
|
} // namespace tensorrt_llm::layers
|