kanshan/TensorRT-LLMs
1
0
Fork 0
mirror of https://github.com/NVIDIA/TensorRT-LLM.git synced 2026-01-14 06:27:45 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity
c89653021e
TensorRT-LLMs/cpp/tensorrt_llm/plugins/gptAttentionPlugin/gptAttentionPlugin.cpp
Kaiyu Xie c89653021e
Update TensorRT-LLM (20240116) (#891) 
* Update TensorRT-LLM

---------

Co-authored-by: Eddie-Wang1120 <81598289+Eddie-Wang1120@users.noreply.github.com>
Co-authored-by: Shixiaowei02 <39303645+Shixiaowei02@users.noreply.github.com>
2024-01-16 20:03:11 +08:00

749 lines
35 KiB
C++
Raw Blame History

/*
* SPDX-FileCopyrightText: Copyright (c) 1993-2022 NVIDIA CORPORATION &
* AFFILIATES. All rights reserved. SPDX-License-Identifier: Apache-2.0
*
* 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 "gptAttentionPlugin.h"
#include "tensorrt_llm/kernels/decoderMaskedMultiheadAttention.h"
#include "tensorrt_llm/kernels/gptKernels.h"
#include "tensorrt_llm/kernels/unfusedAttentionKernels.h"
#include "tensorrt_llm/plugins/common/checkMacrosPlugin.h"
#include "tensorrt_llm/plugins/common/plugin.h"
#include "tensorrt_llm/plugins/gptAttentionCommon/gptAttentionCommon.h"
#include "tensorrt_llm/plugins/gptAttentionCommon/gptAttentionCommonImpl.h"
#include "tensorrt_llm/runtime/iBuffer.h"
#include <algorithm>
#include <cstdint>
#include <functional>
#include <numeric>
using namespace nvinfer1;
using namespace tensorrt_llm::kernels;
using namespace tensorrt_llm::common;
using tensorrt_llm::plugins::GPTAttentionPluginCreator;
using tensorrt_llm::plugins::GPTAttentionPlugin;
static const char* GPT_ATTENTION_PLUGIN_VERSION{"1"};
static const char* GPT_ATTENTION_PLUGIN_NAME{"GPTAttention"};
GPTAttentionPlugin::GPTAttentionPlugin(int num_heads, int num_kv_heads, int head_size, int unidirectional,
float q_scaling, tensorrt_llm::kernels::PositionEmbeddingType position_embedding_type,
int rotary_embedding_dim, // for RoPE. 0 for non-RoPE
float rotary_embedding_base, tensorrt_llm::kernels::RotaryScalingType rotary_embedding_scale_type,
float rotary_embedding_scale, int rotary_embedding_max_positions, int tp_size, int tp_rank, // for ALiBi
bool unfuse_qkv_gemm, // for AutoPP
tensorrt_llm::kernels::ContextFMHAType context_fmha_type, bool multi_block_mode, bool enable_xqa,
int kv_cache_quant_mode, bool remove_input_padding, tensorrt_llm::kernels::AttentionMaskType mask_type,
bool paged_kv_cache, int tokens_per_block, nvinfer1::DataType type, int32_t max_context_length,
bool qkv_bias_enabled, bool cross_attention, int max_distance, bool pos_shift_enabled, bool dense_context_fmha,
bool use_paged_context_fmha, bool use_cache, bool is_medusa_enabled)
: GPTAttentionPluginCommon(num_heads, num_kv_heads, head_size, unidirectional, q_scaling, position_embedding_type,
rotary_embedding_dim, rotary_embedding_base, rotary_embedding_scale_type, rotary_embedding_scale,
rotary_embedding_max_positions, tp_size, tp_rank, unfuse_qkv_gemm, context_fmha_type, multi_block_mode,
enable_xqa, kv_cache_quant_mode, remove_input_padding, mask_type, paged_kv_cache, tokens_per_block, type,
max_context_length, qkv_bias_enabled, cross_attention, max_distance, pos_shift_enabled, dense_context_fmha,
use_paged_context_fmha, use_cache, is_medusa_enabled)
{
initEntryIdx();
}
GPTAttentionPlugin::GPTAttentionPlugin(const void* data, size_t length)
: GPTAttentionPluginCommon(data, length)
{
initEntryIdx();
}
bool GPTAttentionPlugin::isEntryUsed(const IdxEntry& entry) const
{
switch (entry)
{
case IdxEntry::QKV_TENSOR: return true;
case IdxEntry::K_TENSOR: return mUnfuseQkvGemm;
case IdxEntry::V_TENSOR: return mUnfuseQkvGemm;
case IdxEntry::SEQUENCE_LENGTH: return useKVCache();
case IdxEntry::HOST_PAST_KEY_VALUE_LENGTHS: return useKVCache();
case IdxEntry::HOST_MAX_ATTENTION_WINDOW: return true;
case IdxEntry::HOST_SINK_TOKEN_LENGTH: return true;
case IdxEntry::CONTEXT_LENGTHS: return true;
case IdxEntry::CACHE_INDIR: return useKVCache();
case IdxEntry::REQUEST_TYPES: return true;
case IdxEntry::KV_CACHE_BLOCK_POINTERS: return useKVCache() && mPagedKVCache;
case IdxEntry::HOST_KV_CACHE_BLOCK_POINTERS: return useKVCache() && mPagedKVCache;
case IdxEntry::PAST_KEY_VALUE: return useKVCache() && !mPagedKVCache;
case IdxEntry::KV_CACHE_QUANTIZATION_SCALE: return useKVCache() && mKVCacheQuantMode.hasKvCacheQuant();
case IdxEntry::KV_CACHE_DEQUANTIZATION_SCALE: return useKVCache() && mKVCacheQuantMode.hasKvCacheQuant();
case IdxEntry::ALIBI_SLOPES: return isALiBi();
case IdxEntry::RELATIVE_ATTENTION_BIAS: return isRelativePosition();
case IdxEntry::CROSS_QKV: return isCrossAttention();
case IdxEntry::CROSS_QKV_LENGTH: return isCrossAttention();
case IdxEntry::ENCODER_INPUT_LENGTH: return isCrossAttention();
case IdxEntry::HOST_CONTEXT_LENGTH: return mRemovePadding;
case IdxEntry::QKV_BIAS_TENSOR: return mQKVBiasEnabled;
case IdxEntry::MEDUSA_PACKED_MASK: return mIsMedusaEnabled;
case IdxEntry::MEDUSA_POSITION_OFFSETS: return mIsMedusaEnabled;
default: return false;
}
}
void GPTAttentionPlugin::initEntryIdx()
{
mEntryIdx.resize(static_cast<size_t>(IdxEntry::ENUM_SIZE));
size_t entryIdx = 0;
for (int i = 0; i < static_cast<size_t>(IdxEntry::ENUM_SIZE); i++)
{
mEntryIdx[i] = entryIdx;
entryIdx += isEntryUsed(static_cast<IdxEntry>(i));
}
}
GPTAttentionPlugin::IndexType GPTAttentionPlugin::getIdx(const IdxEntry& entry) const
{
TLLM_CHECK_WITH_INFO(
isEntryUsed(entry), common::fmtstr("getIdx() should not be used with entry %lu\n", static_cast<size_t>(entry)));
return mEntryIdx[static_cast<size_t>(entry)];
}
// IPluginV2DynamicExt Methods
GPTAttentionPlugin* GPTAttentionPlugin::clone() const noexcept
{
return dynamic_cast<GPTAttentionPlugin*>(this->cloneImpl<GPTAttentionPlugin>());
}
static int getPackedTensorHiddenDimIndex(bool removePadding)
{
return removePadding ? 1 : 2;
}
// NOTE: generation input length might be larger than one in the Medusa mode.
int GPTAttentionPlugin::getGenerationInputSequenceLength(
const nvinfer1::PluginTensorDesc* inputDesc, int32_t localNbSeq, int32_t localNbTokens) const
{
if (mRemovePadding)
{
// [num_tokens, local_hidden_size] where num_tokens = batch_size * generation_input_length
TLLM_CHECK_WITH_INFO(localNbTokens % localNbSeq == 0,
"seq_len should be same for all generation requests, localNbTokens=%d, localNbSeq=%d", localNbTokens,
localNbSeq);
return localNbTokens / localNbSeq;
}
else
{
// We don't have IFB without mRemovePadding, so just take it out from inputDesc
// [batch_size, seq_len, local_hidden_size]
return inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[1];
}
}
// outputs
// output_tensor [batch_size, seq_len, local_hidden_size] or [num_tokens, local_hidden_size]
// present_key_value_pool (optional if mPagedKVCache is false) [batch_size, 2, local_num_kv_heads, max_seq_len,
// head_size]
nvinfer1::DimsExprs GPTAttentionPlugin::getOutputDimensions(
int outputIndex, const nvinfer1::DimsExprs* inputs, int nbInputs, nvinfer1::IExprBuilder& exprBuilder) noexcept
{
TLLM_CHECK(outputIndex == 0 || (!mPagedKVCache && useKVCache() && outputIndex == 1));
if (outputIndex == 0)
{
auto ret = inputs[getIdx(IdxEntry::QKV_TENSOR)];
ret.d[getPackedTensorHiddenDimIndex(mRemovePadding)] = exprBuilder.operation(
DimensionOperation::kPROD, *exprBuilder.constant(mHeadSize), *exprBuilder.constant(mNumHeads));
return ret;
}
return inputs[getIdx(IdxEntry::PAST_KEY_VALUE)];
}
bool GPTAttentionPlugin::supportsFormatCombination(
int pos, const nvinfer1::PluginTensorDesc* inOut, int nbInputs, int nbOutputs) noexcept
{
if (pos == getIdx(IdxEntry::CONTEXT_LENGTHS) || pos == getIdx(IdxEntry::REQUEST_TYPES)
|| pos == getIdx(IdxEntry::HOST_MAX_ATTENTION_WINDOW) || pos == getIdx(IdxEntry::HOST_SINK_TOKEN_LENGTH)
|| (isEntryUsed(IdxEntry::MEDUSA_PACKED_MASK) && pos == getIdx(IdxEntry::MEDUSA_PACKED_MASK))
|| (isEntryUsed(IdxEntry::MEDUSA_POSITION_OFFSETS) && pos == getIdx(IdxEntry::MEDUSA_POSITION_OFFSETS)))
{
return inOut[pos].type == nvinfer1::DataType::kINT32;
}
else if (useKVCache()
&& (pos == getIdx(IdxEntry::SEQUENCE_LENGTH) || pos == getIdx(IdxEntry::HOST_PAST_KEY_VALUE_LENGTHS)
|| pos == getIdx(IdxEntry::CACHE_INDIR)))
{
return inOut[pos].type == nvinfer1::DataType::kINT32;
}
else if (useKVCache() && mKVCacheQuantMode.hasKvCacheQuant()
&& (pos == getIdx(IdxEntry::KV_CACHE_DEQUANTIZATION_SCALE)
|| pos == getIdx(IdxEntry::KV_CACHE_QUANTIZATION_SCALE)))
{
// kv_scale for mType->int8/fp8 and int8/fp8->mType conversion
return inOut[pos].type == nvinfer1::DataType::kFLOAT && inOut[pos].format == TensorFormat::kLINEAR;
}
else if (mPagedKVCache
&& (pos == getIdx(IdxEntry::KV_CACHE_BLOCK_POINTERS) || pos == getIdx(IdxEntry::HOST_KV_CACHE_BLOCK_POINTERS)))
{
// pointers to kv cache blocks
return inOut[pos].type == nvinfer1::DataType::kINT64 && inOut[pos].format == TensorFormat::kLINEAR;
}
else if (mKVCacheQuantMode.hasInt8KvCache()
&& (!mPagedKVCache && (pos == getIdx(IdxEntry::PAST_KEY_VALUE) || pos == nbInputs + 1)))
{
// If use Int8 K/V cache we require I/O KV values to int8
return (inOut[pos].type == nvinfer1::DataType::kINT8) && (inOut[pos].format == TensorFormat::kLINEAR);
}
else if (mKVCacheQuantMode.hasFp8KvCache()
&& (!mPagedKVCache && (pos == getIdx(IdxEntry::PAST_KEY_VALUE) || pos == nbInputs + 1)))
{
// If use FP8 K/V cache we require I/O KV values to FP8
return (inOut[pos].type == nvinfer1::DataType::kFP8) && (inOut[pos].format == TensorFormat::kLINEAR);
}
else if (mRemovePadding && (pos == getIdx(IdxEntry::HOST_CONTEXT_LENGTH)))
{
return inOut[pos].type == nvinfer1::DataType::kINT32 && inOut[pos].format == TensorFormat::kLINEAR;
}
else if (mCrossAttention
&& (pos == getIdx(IdxEntry::CROSS_QKV_LENGTH) || pos == getIdx(IdxEntry::ENCODER_INPUT_LENGTH)))
{
return inOut[pos].type == nvinfer1::DataType::kINT32;
}
else
{
return (inOut[pos].type == mType) && (inOut[pos].format == TensorFormat::kLINEAR);
}
return false;
}
void GPTAttentionPlugin::configurePlugin(const nvinfer1::DynamicPluginTensorDesc* in, int nbInputs,
const nvinfer1::DynamicPluginTensorDesc* out, int nbOutputs) noexcept
{
TLLM_CHECK(mHeadSize > 0);
}
size_t GPTAttentionPlugin::getWorkspaceSize(const nvinfer1::PluginTensorDesc* inputs, int nbInputs,
const nvinfer1::PluginTensorDesc* outputs, int nbOutputs) const noexcept
{
const int max_context_length = mMaxContextLength;
const int cross_qkv_length = isCrossAttention() ? inputs[getIdx(IdxEntry::CROSS_QKV_LENGTH)].dims.d[0] : 0;
const int nbReq = inputs[getIdx(IdxEntry::CONTEXT_LENGTHS)].dims.d[0];
auto const type = inputs[getIdx(IdxEntry::QKV_TENSOR)].type;
const int max_kv_cache_length
= isCrossAttention() ? cross_qkv_length : (useKVCache() ? inputs[getIdx(IdxEntry::CACHE_INDIR)].dims.d[2] : 0);
size_t const context_workspace_size
= getWorkspaceSizeForContext(type, nbReq, max_context_length, max_kv_cache_length, cross_qkv_length);
const int total_num_seq = inputs[getIdx(IdxEntry::CONTEXT_LENGTHS)].dims.d[0];
size_t const generation_workspace_size = getWorkspaceSizeForGeneration(type, total_num_seq, max_kv_cache_length);
size_t attention_input_workspace_size = 0;
if (mUnfuseQkvGemm)
{
int const local_hidden_units_q
= inputs[getIdx(IdxEntry::QKV_TENSOR)].dims.d[getPackedTensorHiddenDimIndex(mRemovePadding)];
int const local_hidden_units_kv
= inputs[getIdx(IdxEntry::K_TENSOR)].dims.d[getPackedTensorHiddenDimIndex(mRemovePadding)];
size_t const size = tensorrt_llm::runtime::BufferDataType(type).getSize();
size_t const attention_input_size
= size * nbReq * max_context_length * (local_hidden_units_q + 2 * local_hidden_units_kv);
size_t workspaces[1];
workspaces[0] = attention_input_size;
attention_input_workspace_size = tensorrt_llm::common::calculateTotalWorkspaceSize(workspaces, 1);
}
return std::max(context_workspace_size, generation_workspace_size) + attention_input_workspace_size;
}
static size_t getStride(nvinfer1::Dims const& dims, int n)
{
TLLM_CHECK(n >= 0 && n < dims.nbDims);
return std::accumulate(dims.d + n + 1, dims.d + dims.nbDims, 1, std::multiplies<size_t>{});
}
template <typename T, typename KVCacheBuffer>
int GPTAttentionPlugin::enqueueImpl(const nvinfer1::PluginTensorDesc* inputDesc,
const nvinfer1::PluginTensorDesc* outputDesc, const void* const* inputs, void* const* outputs, void* workspace,
cudaStream_t stream)
{
int32_t const nbSeq = inputDesc[getIdx(IdxEntry::CONTEXT_LENGTHS)].dims.d[0];
int32_t const beam_width = useKVCache() ? inputDesc[getIdx(IdxEntry::CACHE_INDIR)].dims.d[1] : 1;
RequestType const* reqTypes = static_cast<RequestType const*>(inputs[getIdx(IdxEntry::REQUEST_TYPES)]);
int32_t nbContextRequests = 0;
int32_t contextTokenIdxEnd = 0;
// count context requests
for (int32_t seqIdx = 0; seqIdx < nbSeq; seqIdx++)
{
if (reqTypes[seqIdx] != RequestType::kCONTEXT)
{
break;
}
++nbContextRequests;
contextTokenIdxEnd += mRemovePadding
? static_cast<int32_t const*>(inputs[getIdx(IdxEntry::HOST_CONTEXT_LENGTH)])[seqIdx]
: inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[1];
}
for (int32_t seqIdx = nbContextRequests; seqIdx < nbSeq; seqIdx++)
{
TLLM_CHECK(reqTypes[seqIdx] == RequestType::kGENERATION);
}
// mixed requests require mRemovePadding and mPagedKVCache
if (nbContextRequests != 0 && nbContextRequests != nbSeq)
{
TLLM_CHECK(mRemovePadding && mPagedKVCache);
}
if (nbContextRequests > 0)
{
auto seqIdxBeg = 0;
auto tokenIdxBeg = 0;
auto localNbTokens = contextTokenIdxEnd;
enqueueSome<T, KVCacheBuffer>(seqIdxBeg, nbContextRequests, tokenIdxBeg, localNbTokens, inputDesc, outputDesc,
inputs, outputs, workspace, stream);
}
if (auto nbGenerationSeq = nbSeq - nbContextRequests; nbGenerationSeq > 0)
{
auto seqIdxBeg = nbContextRequests;
auto tokenIdxBeg = contextTokenIdxEnd;
// if mRemovePadding is true, we may have IFB, and need to remove context tokens.
// if mRemovePadding is false, it is only generation requests, so just multiply batch_beam and seq_len (May not
// 1 for Parallel Decoding)
auto localNbTokens = mRemovePadding
? inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[0] - contextTokenIdxEnd
: inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[0] * inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[1];
enqueueSome<T, KVCacheBuffer>(seqIdxBeg, nbGenerationSeq, tokenIdxBeg, localNbTokens, inputDesc, outputDesc,
inputs, outputs, workspace, stream);
}
return 0;
}
template <typename T, typename KVCacheBuffer>
int GPTAttentionPlugin::enqueueSome(int32_t seqIdxBeg, int32_t localNbSeq, int32_t tokenIdxBeg, int32_t localNbTokens,
const nvinfer1::PluginTensorDesc* inputDesc, const nvinfer1::PluginTensorDesc* outputDesc,
const void* const* inputs, void* const* outputs, void* workspace, cudaStream_t stream)
{
// relative_attention_bias [head_num, max_seq_len, max_seq_len] (optional in relative position)
// or [head_num, num_buckets] (optional in implicit relative attention)
// cross_qkv [batch_size, seq_len, 3 * local_hidden_size] or [num_tokens, 3 * local_hidden_size]
// when enable remove_input_padding (optional in cross attention mode)
// cross_qkv_length [int] max encoder input context length (optional in cross attention mode)
// encoder_input_lengths [batch_size] raw sequence lengths (optional in cross attention mode)
const T* attention_input = static_cast<const T*>(inputs[getIdx(IdxEntry::QKV_TENSOR)])
+ inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[getPackedTensorHiddenDimIndex(mRemovePadding)]
* size_t(tokenIdxBeg);
const T* qkv_bias = nullptr;
if (mQKVBiasEnabled)
{
qkv_bias = reinterpret_cast<const T*>(inputs[getIdx(IdxEntry::QKV_BIAS_TENSOR)]);
}
auto const reqTypeInBatchPtr = static_cast<RequestType const*>(inputs[getIdx(IdxEntry::REQUEST_TYPES)]) + seqIdxBeg;
bool const is_context = (reqTypeInBatchPtr[0] == RequestType::kCONTEXT);
if (mUnfuseQkvGemm)
{
int const max_seqlen = inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[mRemovePadding ? 0 : 1];
int const batch_size = mRemovePadding ? 1 : inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[0];
T const* attention_input_q = static_cast<T const*>(inputs[getIdx(IdxEntry::QKV_TENSOR)]);
T const* attention_input_k = static_cast<T const*>(inputs[getIdx(IdxEntry::K_TENSOR)]);
T const* attention_input_v = static_cast<T const*>(inputs[getIdx(IdxEntry::V_TENSOR)]);
size_t const hidden_units_q
= inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[getPackedTensorHiddenDimIndex(mRemovePadding)];
size_t const hidden_units_kv
= inputDesc[getIdx(IdxEntry::K_TENSOR)].dims.d[getPackedTensorHiddenDimIndex(mRemovePadding)];
size_t const hidden_units = hidden_units_q + 2 * hidden_units_kv;
size_t const size_qkv = sizeof(T) * hidden_units;
size_t const size_q = sizeof(T) * hidden_units_q;
size_t const size_kv = sizeof(T) * hidden_units_kv;
size_t const total_size = size_qkv * batch_size * max_seqlen;
int8_t* workspace_byte_ptr = reinterpret_cast<int8_t*>(workspace);
size_t offset = 0;
T* attention_input_qkv = reinterpret_cast<T*>(nextWorkspacePtr(workspace_byte_ptr, offset, total_size));
workspace = reinterpret_cast<void*>(workspace_byte_ptr + offset);
cudaMemcpy2DAsync(attention_input_qkv, size_qkv, attention_input_q, size_q, size_q, batch_size * max_seqlen,
cudaMemcpyDeviceToDevice, stream);
cudaMemcpy2DAsync(attention_input_qkv + hidden_units_q, size_qkv, attention_input_k, size_kv, size_kv,
batch_size * max_seqlen, cudaMemcpyDeviceToDevice, stream);
cudaMemcpy2DAsync(attention_input_qkv + hidden_units_q + hidden_units_kv, size_qkv, attention_input_v, size_kv,
size_kv, batch_size * max_seqlen, cudaMemcpyDeviceToDevice, stream);
attention_input = attention_input_qkv + hidden_units * tokenIdxBeg;
}
const int* context_q_lengths = reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::CONTEXT_LENGTHS)]) + seqIdxBeg;
const int* sequence_kv_length = useKVCache()
? static_cast<const int*>(inputs[getIdx(IdxEntry::SEQUENCE_LENGTH)]) + seqIdxBeg
: context_q_lengths;
// Note we still need context length during generation for MMHA optimization.
int32_t const max_context_q_len = [&]()
{
if (!mRemovePadding)
{
return inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims.d[1];
}
auto const host_context_lengths
= static_cast<int32_t const*>(inputs[getIdx(IdxEntry::HOST_CONTEXT_LENGTH)]) + seqIdxBeg;
return *std::max_element(host_context_lengths, host_context_lengths + localNbSeq);
}();
TLLM_CHECK(max_context_q_len <= mMaxContextLength);
int max_encoder_context_len = isCrossAttention() ? inputDesc[getIdx(IdxEntry::CROSS_QKV_LENGTH)].dims.d[0] : 0;
// for enc-dec model, since decoder_input_ids could be longer than 1,
// such model has an encoder context (for cross attn) and an decoder context (for self attn)
// clarify 3 lens:
// -- max_context_q_len: len of decoder input. No "max" concept, it's what it is given.
// Also called (decoder_)input_seq_length, normally 1 for encoder-decoder start token
// -- max_seq_len: max allowed len of decoder output, i.e. final results
// -- max_encoder_context_len: len of encoder input (in cross attn). Also called encoder_input_seq_length
const int beamWidth = useKVCache() ? inputDesc[getIdx(IdxEntry::CACHE_INDIR)].dims.d[1] : 1;
// Commonly, cyclic_attention_window_size, and max_attention_window_size will be the same
// unless each layer has different attention window sizes.
// the kv_cache capacity.
const int max_attention_window_size = isCrossAttention()
? max_encoder_context_len
: (useKVCache() ? inputDesc[getIdx(IdxEntry::CACHE_INDIR)].dims.d[2] : 0);
// The cyclic_attention_window_size will determine the cyclic kv cache position of new tokens.
// Note that this cyclic_attention_window_size might be smaller than the actual kv cache capactity.
const int cyclic_attention_window_size = isCrossAttention()
? max_encoder_context_len
: reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::HOST_MAX_ATTENTION_WINDOW)])[0];
const int sink_token_length = reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::HOST_SINK_TOKEN_LENGTH)])[0];
const float* kv_scale_orig_quant = nullptr;
const float* kv_scale_quant_orig = nullptr;
if (useKVCache() && mKVCacheQuantMode.hasKvCacheQuant())
{
assert(inputDesc[getIdx(IdxEntry::KV_CACHE_QUANTIZATION_SCALE)].type == nvinfer1::DataType::kFLOAT);
assert(inputDesc[getIdx(IdxEntry::KV_CACHE_DEQUANTIZATION_SCALE)].type == nvinfer1::DataType::kFLOAT);
kv_scale_orig_quant = reinterpret_cast<const float*>(inputs[getIdx(IdxEntry::KV_CACHE_QUANTIZATION_SCALE)]);
kv_scale_quant_orig = reinterpret_cast<const float*>(inputs[getIdx(IdxEntry::KV_CACHE_DEQUANTIZATION_SCALE)]);
}
int max_blocks_per_sequence = 0;
void* block_pointers = nullptr;
void* host_block_pointers = nullptr;
if (useKVCache() && mPagedKVCache)
{
auto& kvCacheBlockPointers = inputDesc[getIdx(IdxEntry::KV_CACHE_BLOCK_POINTERS)];
auto& kvCacheBlockPointersShape = inputDesc[getIdx(IdxEntry::KV_CACHE_BLOCK_POINTERS)].dims;
max_blocks_per_sequence = kvCacheBlockPointersShape.d[kvCacheBlockPointersShape.nbDims - 1];
auto offset = getStride(kvCacheBlockPointersShape, 0) * seqIdxBeg;
auto const typed_block_pointers
= static_cast<void* const*>(inputs[getIdx(IdxEntry::KV_CACHE_BLOCK_POINTERS)]) + offset;
block_pointers = const_cast<void*>(static_cast<void const*>(typed_block_pointers));
auto const typed_host_block_pointers
= static_cast<void* const*>(inputs[getIdx(IdxEntry::HOST_KV_CACHE_BLOCK_POINTERS)]) + offset;
host_block_pointers = const_cast<void*>(static_cast<void const*>(typed_host_block_pointers));
}
T* context_buf_
= (T*) (outputs[0]) + outputDesc[0].dims.d[getPackedTensorHiddenDimIndex(mRemovePadding)] * tokenIdxBeg;
void* key_value_cache = nullptr;
if (useKVCache() && !mPagedKVCache)
{
auto const cacheElemSize = (mKVCacheQuantMode.hasKvCacheQuant() ? 1 : sizeof(T));
key_value_cache
= static_cast<std::byte*>(outputs[1]) + cacheElemSize * getStride(outputDesc[1].dims, 0) * seqIdxBeg;
void const* past_key_value_cache = inputs[getIdx(IdxEntry::PAST_KEY_VALUE)];
if (past_key_value_cache != outputs[1])
{
auto shape = outputDesc[1].dims;
auto const size = std::accumulate(shape.d, shape.d + shape.nbDims, 1, std::multiplies<size_t>{});
cudaMemcpyAsync(outputs[1], past_key_value_cache, size, cudaMemcpyDeviceToDevice, stream);
}
}
const T* alibi_slopes = isALiBi() ? static_cast<const T*>(inputs[getIdx(IdxEntry::ALIBI_SLOPES)]) : nullptr;
const int* medusa_packed_mask = nullptr;
const int* medusa_position_offsets = nullptr;
int num_medusa_tokens = 0;
if (mIsMedusaEnabled)
{
// First dimension of medusa_packed_mask is num_medusa_tokens + 1.
num_medusa_tokens = inputDesc[getIdx(IdxEntry::MEDUSA_PACKED_MASK)].dims.d[0] - 1;
if (num_medusa_tokens > 0)
{
medusa_packed_mask = static_cast<const int*>(inputs[getIdx(IdxEntry::MEDUSA_PACKED_MASK)]);
medusa_position_offsets = static_cast<const int*>(inputs[getIdx(IdxEntry::MEDUSA_POSITION_OFFSETS)]);
}
}
int32_t const* max_context_kv_len_list = useKVCache()
? static_cast<const int*>(inputs[getIdx(IdxEntry::HOST_PAST_KEY_VALUE_LENGTHS)]) + seqIdxBeg
: nullptr;
int32_t const max_context_kv_len = useKVCache()
? *std::max_element(max_context_kv_len_list, max_context_kv_len_list + localNbSeq)
: max_context_q_len;
if (is_context) // context stage
{
const int batch_size = localNbSeq;
const int request_batch_size = batch_size;
// num of total tokens (without paddings when remove paddings).
int num_encoder_tokens = 0;
if (isCrossAttention())
{
if (!mRemovePadding)
{
num_encoder_tokens = request_batch_size * max_encoder_context_len;
}
else
{
num_encoder_tokens = inputDesc[getIdx(IdxEntry::CROSS_QKV)].dims.d[0];
}
}
EnqueueContextParams<T, KVCacheBuffer> enqueue_params{attention_input, qkv_bias, max_context_q_len,
max_context_kv_len, max_attention_window_size, cyclic_attention_window_size, sink_token_length,
context_q_lengths, sequence_kv_length, kv_scale_orig_quant, kv_scale_quant_orig, alibi_slopes, context_buf_,
key_value_cache, block_pointers, host_block_pointers, batch_size, localNbTokens, max_blocks_per_sequence,
workspace};
if (isRelativePosition())
{
enqueue_params.relative_attention_bias
= static_cast<const T*>(inputs[getIdx(IdxEntry::RELATIVE_ATTENTION_BIAS)]);
enqueue_params.relative_attention_bias_stride
= inputDesc[getIdx(IdxEntry::RELATIVE_ATTENTION_BIAS)].dims.d[1]; // max_seq_len or num_buckets
}
if (isCrossAttention())
{
enqueue_params.cross_qkv = static_cast<const T*>(inputs[getIdx(IdxEntry::CROSS_QKV)]);
enqueue_params.cross_qkv_length = max_encoder_context_len;
enqueue_params.encoder_input_lengths
= reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::ENCODER_INPUT_LENGTH)]) + seqIdxBeg;
enqueue_params.num_encoder_tokens = num_encoder_tokens;
}
enqueueContext<T, KVCacheBuffer>(enqueue_params, stream);
}
else // generation stage; max_context_q_len == input_seq_len == 1
{
TLLM_CHECK_WITH_INFO(useKVCache(), "KV-cache-less is only supported for context");
int batch_beam = localNbSeq;
TLLM_CHECK(batch_beam % beamWidth == 0);
int32_t const num_requests = batch_beam / beamWidth;
const int* cache_indir
= beamWidth == 1 ? nullptr : reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::CACHE_INDIR)]);
const int* host_context_lengths
= mRemovePadding ? reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::HOST_CONTEXT_LENGTH)]) : nullptr;
const int input_seq_length = getGenerationInputSequenceLength(inputDesc, localNbSeq, localNbTokens);
auto qkvDims = inputDesc[getIdx(IdxEntry::QKV_TENSOR)].dims;
TLLM_CHECK_WITH_INFO(input_seq_length == 1 || mIsMedusaEnabled,
"Only Medusa mode supports input length > 1 in the generation phase, input_seq_length=%d, "
"mIsMedusaEnabled=%s, nDims=%d, (%d, %d, %d)",
input_seq_length, mIsMedusaEnabled ? "true" : "false", qkvDims.nbDims, qkvDims.d[0], qkvDims.d[1],
qkvDims.d[2]);
TLLM_CHECK_WITH_INFO(input_seq_length == num_medusa_tokens + 1, "The generation input length is not expected.");
EnqueueGenerationParams<T, KVCacheBuffer> enqueue_params{attention_input, qkv_bias, input_seq_length,
sequence_kv_length, max_context_kv_len, beamWidth, context_q_lengths, kv_scale_orig_quant,
kv_scale_quant_orig, alibi_slopes, context_buf_, key_value_cache, block_pointers, max_attention_window_size,
cyclic_attention_window_size, sink_token_length, num_requests, max_blocks_per_sequence, cache_indir,
workspace, max_context_kv_len_list};
enqueue_params.host_context_lengths = host_context_lengths;
if (isRelativePosition())
{
enqueue_params.relative_attention_bias
= static_cast<const T*>(inputs[getIdx(IdxEntry::RELATIVE_ATTENTION_BIAS)]);
enqueue_params.relative_attention_bias_stride
= inputDesc[getIdx(IdxEntry::RELATIVE_ATTENTION_BIAS)].dims.d[1]; // max_seq_len or num_buckets
}
if (isCrossAttention())
{
enqueue_params.encoder_input_lengths
= reinterpret_cast<const int*>(inputs[getIdx(IdxEntry::ENCODER_INPUT_LENGTH)]) + seqIdxBeg;
}
if (mIsMedusaEnabled)
{
enqueue_params.medusa_packed_mask = medusa_packed_mask;
enqueue_params.medusa_position_offsets = medusa_position_offsets;
}
enqueueGeneration<T, KVCacheBuffer>(enqueue_params, stream);
}
return 0;
}
template <typename T>
int GPTAttentionPlugin::enqueueDispatchKVCacheType(const nvinfer1::PluginTensorDesc* inputDesc,
const nvinfer1::PluginTensorDesc* outputDesc, const void* const* inputs, void* const* outputs, void* workspace,
cudaStream_t stream)
{
if (mPagedKVCache)
{
return enqueueImpl<T, KVBlockArray>(inputDesc, outputDesc, inputs, outputs, workspace, stream);
}
else
{
return enqueueImpl<T, KVLinearBuffer>(inputDesc, outputDesc, inputs, outputs, workspace, stream);
}
return 0;
}
int GPTAttentionPlugin::enqueue(const nvinfer1::PluginTensorDesc* inputDesc,
const nvinfer1::PluginTensorDesc* outputDesc, const void* const* inputs, void* const* outputs, void* workspace,
cudaStream_t stream) noexcept
{
if (mType == nvinfer1::DataType::kHALF)
{
return enqueueDispatchKVCacheType<half>(inputDesc, outputDesc, inputs, outputs, workspace, stream);
}
else if (mType == nvinfer1::DataType::kFLOAT)
{
return enqueueDispatchKVCacheType<float>(inputDesc, outputDesc, inputs, outputs, workspace, stream);
}
#ifdef ENABLE_BF16
else if (mType == nvinfer1::DataType::kBF16)
{
return enqueueDispatchKVCacheType<__nv_bfloat16>(inputDesc, outputDesc, inputs, outputs, workspace, stream);
}
#endif
return 0;
}
// IPluginV2Ext Methods
nvinfer1::DataType GPTAttentionPlugin::getOutputDataType(
int index, const nvinfer1::DataType* inputTypes, int nbInputs) const noexcept
{
TLLM_CHECK(index == 0 || (!mPagedKVCache && index == 1));
if (index == 0)
{
return inputTypes[getIdx(IdxEntry::QKV_TENSOR)];
}
else
{
return inputTypes[getIdx(IdxEntry::PAST_KEY_VALUE)];
}
}
// IPluginV2 Methods
const char* GPTAttentionPlugin::getPluginType() const noexcept
{
return GPT_ATTENTION_PLUGIN_NAME;
}
const char* GPTAttentionPlugin::getPluginVersion() const noexcept
{
return GPT_ATTENTION_PLUGIN_VERSION;
}
int GPTAttentionPlugin::getNbOutputs() const noexcept
{
return (mPagedKVCache || !useKVCache()) ? 1 : 2;
}
size_t GPTAttentionPlugin::getSerializationSize() const noexcept
{
return GPTAttentionPluginCommon::getCommonSerializationSize();
}
void GPTAttentionPlugin::serialize(void* buffer) const noexcept
{
GPTAttentionPluginCommon::serializeCommon(buffer);
}
///////////////
GPTAttentionPluginCreator::GPTAttentionPluginCreator()
: GPTAttentionPluginCreatorCommon()
{
mPluginAttributes.emplace_back(PluginField("in_flight_batching", nullptr, PluginFieldType::kINT8, 0));
mFC.nbFields = mPluginAttributes.size();
mFC.fields = mPluginAttributes.data();
}
const char* GPTAttentionPluginCreator::getPluginName() const noexcept
{
return GPT_ATTENTION_PLUGIN_NAME;
}
const char* GPTAttentionPluginCreator::getPluginVersion() const noexcept
{
return GPT_ATTENTION_PLUGIN_VERSION;
}
const PluginFieldCollection* GPTAttentionPluginCreator::getFieldNames() noexcept
{
return &mFC;
}
IPluginV2* GPTAttentionPluginCreator::createPlugin(const char* name, const PluginFieldCollection* fc) noexcept
{
PluginFieldParser p{fc->nbFields, fc->fields};
try
{
auto* obj = new GPTAttentionPlugin(p.getScalar<int32_t>("num_heads").value(),
p.getScalar<int32_t>("num_kv_heads").value(), p.getScalar<int32_t>("head_size").value(),
p.getScalar<int32_t>("unidirectional").value(), p.getScalar<float>("q_scaling").value(),
static_cast<PositionEmbeddingType>(p.getScalar<int8_t>("position_embedding_type").value()),
p.getScalar<int32_t>("rotary_embedding_dim").value(), p.getScalar<float>("rotary_embedding_base").value(),
static_cast<RotaryScalingType>(p.getScalar<int8_t>("rotary_embedding_scale_type").value()),
p.getScalar<float>("rotary_embedding_scale").value(),
p.getScalar<int32_t>("rotary_embedding_max_positions").value(),
static_cast<int32_t>(p.getScalar<int32_t>("tp_size").value()),
static_cast<int32_t>(p.getScalar<int32_t>("tp_rank").value()),
static_cast<bool>(p.getScalar<int8_t>("unfuse_qkv_gemm").value()),
static_cast<ContextFMHAType>(p.getScalar<int8_t>("context_fmha_type").value()),
static_cast<bool>(p.getScalar<int8_t>("multi_block_mode").value()),
static_cast<bool>(p.getScalar<int8_t>("enable_xqa").value()),
p.getScalar<int32_t>("kv_cache_quant_mode").value(),
static_cast<bool>(p.getScalar<int8_t>("remove_input_padding").value()),
static_cast<AttentionMaskType>(p.getScalar<int32_t>("mask_type").value()),
static_cast<bool>(p.getScalar<int32_t>("paged_kv_cache").value()),
p.getScalar<int32_t>("tokens_per_block").value(),
static_cast<nvinfer1::DataType>(p.getScalar<int32_t>("type_id").value()),
p.getScalar<int32_t>("max_context_length").value(),
static_cast<bool>(p.getScalar<int8_t>("qkv_bias_enabled").value()),
static_cast<bool>(p.getScalar<int8_t>("do_cross_attention").value()),
static_cast<int32_t>(p.getScalar<int32_t>("max_distance").value()),
static_cast<bool>(p.getScalar<int8_t>("pos_shift_enabled").value()),
static_cast<bool>(p.getScalar<int8_t>("dense_context_fmha").value()),
static_cast<bool>(p.getScalar<int8_t>("use_paged_context_fmha").value()),
static_cast<bool>(p.getScalar<int32_t>("use_cache").value()),
static_cast<bool>(p.getScalar<int8_t>("is_medusa_enabled").value()));
obj->setPluginNamespace(mNamespace.c_str());
return obj;
}
catch (const std::exception& e)
{
caughtError(e);
}
return nullptr;
}
IPluginV2* GPTAttentionPluginCreator::deserializePlugin(
const char* name, const void* serialData, size_t serialLength) noexcept
{
// This object will be deleted when the network is destroyed, which will
// call GPTAttentionPlugin::destroy()
try
{
auto* obj = new GPTAttentionPlugin(serialData, serialLength);
obj->setPluginNamespace(mNamespace.c_str());
return obj;
}
catch (const std::exception& e)
{
caughtError(e);
}
return nullptr;
}
Reference in New Issue View Git Blame Copy Permalink