TensorRT-LLMs/cpp/tensorrt_llm/runtime/gptDecoderBatched.cpp
Robin Kobus 3de02582dd
refactor: Separate DecoderState from GptDecoderBatched (#4700)
Signed-off-by: Robin Kobus <19427718+Funatiq@users.noreply.github.com>
2025-06-03 09:42:01 +02:00

338 lines
13 KiB
C++

/*
* Copyright (c) 2022-2024, 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.
*/
#include "tensorrt_llm/runtime/gptDecoderBatched.h"
#include "common.h"
#include "decoderState.h"
#include "iBuffer.h"
#include "tensorrt_llm/batch_manager/createNewDecoderRequests.h"
#include "tensorrt_llm/batch_manager/llmRequest.h"
#include "tensorrt_llm/common/assert.h"
#include "tensorrt_llm/executor/types.h"
#include "tensorrt_llm/kernels/decodingKernels.h"
#include "tensorrt_llm/runtime/bufferManager.h"
#include "tensorrt_llm/runtime/cudaEvent.h"
#include <algorithm>
#include <cassert>
#include <memory>
#include <vector>
using namespace tensorrt_llm::runtime;
using TensorPtr = ITensor::SharedPtr;
GptDecoderBatched::GptDecoderBatched(GptDecoderBatched::CudaStreamPtr stream)
: mRuntimeStream{std::move(stream)}
, mBufferManager{mRuntimeStream}
{
}
void GptDecoderBatched::disableLookahead(RequestVector const& genRequests, TensorPtr const& batchSlots)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
std::vector<SamplingConfig> samplingConfigs;
samplingConfigs.reserve(genRequests.size());
auto batchSlotsRange = BufferRange<SizeType32>(*batchSlots);
SizeType32 batchIdx = 0;
for (auto const& llmReq : genRequests)
{
samplingConfigs.push_back(llmReq->mSamplingConfig);
batchSlotsRange[batchIdx] = llmReq->mSeqSlot.value();
batchIdx += 1;
}
auto const batchSize = batchIdx;
std::optional<SamplingConfig> samplingConfig;
if (batchSize > 0)
{
samplingConfig = SamplingConfig(samplingConfigs);
}
TensorPtr batchSlotsView = ITensor::slice(batchSlots, 0, batchSize);
mDecoder->disableLookahead(samplingConfig, batchSize, batchSlots);
CudaEvent event{};
mDecoderStream->record(event);
mRuntimeStream->wait(event);
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
void GptDecoderBatched::setup(executor::DecodingMode const& mode, SizeType32 maxBatchSize, SizeType32 maxBeamWidth,
SizeType32 maxSequenceLength, nvinfer1::DataType dtype, ModelConfig const& modelConfig,
WorldConfig const& worldConfig)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
TLLM_CHECK(maxBatchSize > 0);
TLLM_CHECK(maxBeamWidth > 0);
TLLM_CHECK(maxSequenceLength > 0);
std::shared_ptr<SpeculativeDecodingModule const> speculativeDecodingModulePtr = nullptr;
if (modelConfig.getSpeculativeDecodingMode().predictsDraftTokens())
{
speculativeDecodingModulePtr = modelConfig.getSpeculativeDecodingModulePtr();
}
auto const device = mRuntimeStream->getDevice();
mDecoderStream = std::make_shared<CudaStream>();
TLLM_CHECK(mDecoderStream->getDevice() == device);
auto const vocabSize = modelConfig.getVocabSize();
auto const vocabSizePadded = modelConfig.getVocabSizePadded(worldConfig.getSize());
mDecoder = IGptDecoder::create(mode, dtype, maxBatchSize, maxBeamWidth, vocabSize, vocabSizePadded,
maxSequenceLength, mDecoderStream, speculativeDecodingModulePtr);
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
namespace
{
//! @brief Sets inputs for explicit draft tokens.
void setExplicitDraftTokensInputs(DecodingInput& dInput, decoder_batch::Input const& input)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
dInput.explicitDraftTokensInputs = DecodingInput::ExplicitDraftTokensInputs();
TLLM_CHECK(input.explicitDraftTokensInputs.has_value());
TLLM_CHECK(input.explicitDraftTokensLastInputs.has_value());
dInput.explicitDraftTokensInputs->nextDraftTokens = input.explicitDraftTokensInputs->nextDraftTokens;
dInput.explicitDraftTokensInputs->nextFlatTokens = input.explicitDraftTokensInputs->nextFlatTokens;
dInput.explicitDraftTokensInputs->nextDraftIndices = input.explicitDraftTokensInputs->nextDraftIndices;
dInput.explicitDraftTokensInputs->nextDraftProbs = input.explicitDraftTokensInputs->nextDraftProbs;
dInput.explicitDraftTokensInputs->lastDraftTokens = input.explicitDraftTokensLastInputs->draftTokens;
dInput.explicitDraftTokensInputs->lastDraftIndices = input.explicitDraftTokensLastInputs->draftIndices;
dInput.explicitDraftTokensInputs->lastPositionIdsBase = input.explicitDraftTokensLastInputs->positionIdsBase;
dInput.explicitDraftTokensInputs->masks = input.explicitDraftTokensInputs->masks;
dInput.explicitDraftTokensInputs->packedPositionIds = input.explicitDraftTokensInputs->packedPositionIds;
dInput.explicitDraftTokensInputs->bestPathLengths = input.explicitDraftTokensInputs->bestPathLengths;
dInput.explicitDraftTokensInputs->bestPathIndices = input.explicitDraftTokensInputs->bestPathIndices;
dInput.explicitDraftTokensInputs->nextGenerationLengths = input.explicitDraftTokensInputs->nextGenerationLengths;
dInput.explicitDraftTokensInputs->lastGenerationLengths = input.explicitDraftTokensLastInputs->generationLengths;
dInput.explicitDraftTokensInputs->maxGenLengthDevice = input.explicitDraftTokensInputs->maxGenToken;
dInput.explicitDraftTokensInputs->seqSlots = input.batchSlotsRequestOrder;
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
//! @brief Sets inputs for eagle decoding.
void setEagleInputs(DecodingInput& dInput, decoder_batch::Input const& input)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
TLLM_CHECK(input.eagleInputs.has_value());
TLLM_CHECK(input.eagleLastInputs.has_value());
dInput.eagleInputs = DecodingInput::EagleInputs(input.eagleInputs->nextDraftTokens,
input.eagleInputs->nextDraftLens, input.eagleInputs->nextDraftPaths, input.eagleLastInputs->draftTokens,
input.eagleLastInputs->draftLens, input.eagleLastInputs->draftPaths, input.eagleInputs->acceptedTokens,
input.eagleInputs->acceptedLens, input.eagleInputs->acceptedPaths, input.eagleInputs->chunkedContextNextTokens,
input.batchSlotsRequestOrder);
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
//! @brief Prepare Input and Output for decoder step.
// TODO: produce new input and output objects
void prepareForward(decoder::DecoderState const& decoderState, SizeType32 step, decoder_batch::Output& output,
decoder_batch::Input const& input, BufferManager const& bufferManager)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
auto const maxBeamWidth = decoderState.getMaxBeamWidth();
auto const speculativeDecodingMode = decoderState.getSpeculativeDecodingMode();
auto& dInput = decoderState.getJointDecodingInput();
auto& dOutput = decoderState.getJointDecodingOutput();
if (maxBeamWidth > 1)
{
dInput.cacheIndirection = input.cacheIndirection;
dOutput.cacheIndirection = output.cacheIndirection;
dInput.generationSteps = input.generationSteps; // For Variable-Beam-Width-Search
}
if (speculativeDecodingMode.isExplicitDraftTokens())
{
setExplicitDraftTokensInputs(dInput, input);
}
else if (speculativeDecodingMode.isEagle())
{
setEagleInputs(dInput, input);
}
dInput.batchSlots = input.batchSlots.at(step);
dInput.batchSize = static_cast<SizeType32>(dInput.batchSlots->getSize());
dInput.logitsVec = input.logits.at(step);
TensorPtr finishedStepsInput = ITensor::slice(decoderState.getFinishedSteps(), step, 1);
TensorPtr finishedStepsOutput
= ITensor::slice(decoderState.getFinishedSteps(), std::min(input.maxDecoderSteps - 1, step + 1), 1);
finishedStepsInput->squeeze(0);
finishedStepsOutput->squeeze(0);
TensorPtr newTokensStepView
= ITensor::slice(dOutput.newTokensSteps, step, decoderState.getMaxDecodingDecoderTokens());
dInput.finishReasons = finishedStepsInput;
if (speculativeDecodingMode.isMedusa())
{
dInput.medusaInputs->medusaLogits = input.predictedDraftLogits;
}
if (speculativeDecodingMode.isDraftTokensExternal())
{
dInput.externalDraftTokensInputs->step = step;
// WAR: reset finished state for generation requests
if (step == 0)
{
auto batchSlotsRange = BufferRange<SizeType32 const>(*dInput.batchSlots);
for (auto batchSlot : batchSlotsRange)
{
TensorPtr finishedSteps = ITensor::slice(finishedStepsInput, batchSlot, 1);
bufferManager.setZero(*finishedSteps);
}
}
}
dOutput.newTokens = newTokensStepView;
dOutput.finishReasons = finishedStepsOutput;
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
} // namespace
void GptDecoderBatched::forwardDispatch(
decoder::DecoderState const& decoderState, decoder_batch::Output& output, decoder_batch::Input const& input)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
for (SizeType32 step = 0; step < input.maxDecoderSteps; ++step)
{
BufferManager manager{mDecoderStream};
prepareForward(decoderState, step, output, input, manager);
if (decoderState.getJointDecodingInput().batchSize > 0)
{
mDecoder->forwardAsync(decoderState.getJointDecodingOutput(), decoderState.getJointDecodingInput());
}
}
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
CudaEvent GptDecoderBatched::forwardAsync(
decoder::DecoderState const& decoderState, decoder_batch::Output& output, decoder_batch::Input const& input)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
auto eventStart = CudaEvent{};
mRuntimeStream->record(eventStart);
mDecoderStream->wait(eventStart.get());
forwardDispatch(decoderState, output, input);
CudaEvent event{};
mDecoderStream->record(event);
mRuntimeStream->wait(event);
CudaEvent eventStop{};
mRuntimeStream->record(eventStop);
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
return eventStop;
}
void GptDecoderBatched::forward(
decoder::DecoderState const& decoderState, decoder_batch::Output& output, decoder_batch::Input const& input)
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
auto decoderFinishEvent = forwardAsync(decoderState, output, input);
decoderFinishEvent.synchronize();
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
}
namespace
{
std::pair<DecodingInput, DecodingOutput> prepareGatherTree(
decoder::DecoderState const& decoderState, SizeType32 batchSlot, bool streaming, CudaStream const& stream)
{
auto& dJointInput = decoderState.getJointDecodingInput();
auto& dJointOutput = decoderState.getJointDecodingOutput();
auto slice = [batchSlot](auto& a, auto const& b)
{
if (b && b->getShape().d[0] > 0)
{
a = ITensor::slice(b, batchSlot, 1);
}
};
// Prepare a slice of dJointInput and dJointOutput for gatherTree
DecodingInput dInput{dJointInput};
slice(dInput.endIds, dJointInput.endIds);
slice(dInput.lengths, dJointInput.lengths);
DecodingOutput dOutput{
ITensor::slice(dJointOutput.ids, batchSlot, 1), ITensor::slice(dJointOutput.gatheredIds, batchSlot, 1)};
dOutput.beamHypotheses = dJointOutput.beamHypotheses.slice(batchSlot, 1);
slice(dOutput.parentIds, dJointOutput.parentIds);
slice(dOutput.cumLogProbs, dJointOutput.cumLogProbs);
slice(dOutput.cacheIndirection, dJointOutput.cacheIndirection);
slice(dOutput.lengths, dJointOutput.lengths);
slice(dOutput.finishReasons, dJointOutput.finishReasons);
slice(dOutput.logProbs, dJointOutput.logProbs);
dOutput.newTokens = ITensor::view(dJointOutput.newTokens);
TLLM_CHECK(dOutput.newTokens->getShape().d[0] == 1);
dOutput.newTokens->squeeze(0);
dOutput.newTokens = ITensor::slice(dOutput.newTokens, batchSlot, 1);
dOutput.logProbsTiled = dJointOutput.logProbsTiled;
if (streaming)
{
// in case of streaming we shouldn't overwrite the data in beamHypotheses, since the beam search kernels expect
// ungathered data but the kernels in gatherTree write in-place.
// Thus, we need to make a copy of the beamHypotheses
auto const& beamSearchBuffers = decoderState.getBeamSearchBuffers();
tensorrt_llm::kernels::invokeCopyBeamHypotheses(dOutput.beamHypotheses, beamSearchBuffers.mOutputBeamHypotheses,
*dOutput.cumLogProbs, *beamSearchBuffers.mCumLogProbsTmp, stream, beamSearchBuffers.mNumSMs);
dOutput.beamHypotheses = beamSearchBuffers.mOutputBeamHypotheses;
dOutput.cumLogProbs = beamSearchBuffers.mCumLogProbsTmp;
}
return {(std::move(dInput)), (std::move(dOutput))};
}
} // namespace
// TODO call this at the end of forward if mFinished[i] changes from false to true?
CudaEvent GptDecoderBatched::finalize(decoder::DecoderState const& decoderState, SizeType32 batchSlot,
SamplingConfig const& samplingConfig, bool streaming) const
{
TLLM_LOG_TRACE("%s start", __PRETTY_FUNCTION__);
auto [dInput, dOutput] = prepareGatherTree(decoderState, batchSlot, streaming, *mRuntimeStream);
kernels::gatherTree(dOutput, dInput, samplingConfig, *mRuntimeStream);
CudaEvent event{};
mRuntimeStream->record(event);
TLLM_LOG_TRACE("%s stop", __PRETTY_FUNCTION__);
return event;
}