-
Notifications
You must be signed in to change notification settings - Fork 1k
/
gptDecoder.h
93 lines (72 loc) · 2.87 KB
/
gptDecoder.h
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
/*
* Copyright (c) 2022-2023, 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/cudaAllocator.h"
#include "tensorrt_llm/runtime/bufferManager.h"
#include "tensorrt_llm/runtime/decodingInput.h"
#include "tensorrt_llm/runtime/decodingOutput.h"
#include "tensorrt_llm/runtime/samplingConfig.h"
#include <cstdint>
#include <memory>
#include <NvInferRuntime.h>
namespace tensorrt_llm
{
namespace layers
{
// Forward declaration
template <typename T>
class DynamicDecodeLayer;
} // namespace layers
namespace runtime
{
class IGptDecoder
{
public:
virtual ~IGptDecoder() = default;
virtual void setup(SamplingConfig const& samplingConfig, size_t batchSize) = 0;
virtual bool forward(DecodingOutput& output, DecodingInput const& input) = 0;
virtual void forwardAsync(DecodingOutput& output, DecodingInput const& input) = 0;
static void gatherTree(ITensor& finalOutputIds, DecodingOutput const& decodingOutput,
DecodingInput const& decodingInput, BufferManager const& manager);
static std::unique_ptr<IGptDecoder> create(
nvinfer1::DataType dtype, size_t vocabSize, size_t vocabSizePadded, BufferManager::CudaStreamPtr const& stream);
};
template <typename T>
class GptDecoder : public virtual IGptDecoder
{
public:
using CudaStreamPtr = BufferManager::CudaStreamPtr;
GptDecoder(size_t vocabSize, size_t vocabSizePadded, CudaStreamPtr const& stream);
void setup(SamplingConfig const& samplingConfig, size_t batchSize) override;
bool forward(DecodingOutput& output, DecodingInput const& input) override;
void forwardAsync(DecodingOutput& output, DecodingInput const& input) override;
private:
BufferManager mManager;
common::CudaAllocator mAllocator;
std::shared_ptr<tensorrt_llm::layers::DynamicDecodeLayer<T>> mDynamicDecodeLayer;
};
inline std::unique_ptr<IGptDecoder> IGptDecoder::create(
nvinfer1::DataType dtype, size_t vocabSize, size_t vocabSizePadded, BufferManager::CudaStreamPtr const& stream)
{
switch (dtype)
{
case nvinfer1::DataType::kFLOAT: return std::make_unique<GptDecoder<float>>(vocabSize, vocabSizePadded, stream);
case nvinfer1::DataType::kHALF: return std::make_unique<GptDecoder<half>>(vocabSize, vocabSizePadded, stream);
default: return nullptr;
}
}
} // namespace runtime
} // namespace tensorrt_llm