"\n# Speed Up Model with Mask\n\n## Introduction\n\nPruning algorithms usually use weight masks to simulate the real pruning. Masks can be used\nto check model performance of a specific pruning (or sparsity), but there is no real speedup.\nSince model speedup is the ultimate goal of model pruning, we try to provide a tool to users\nto convert a model to a smaller one based on user provided masks (the masks come from the\npruning algorithms).\n\nThere are two types of pruning. One is fine-grained pruning, it does not change the shape of weights,\nand input/output tensors. Sparse kernel is required to speedup a fine-grained pruned layer.\nThe other is coarse-grained pruning (e.g., channels), shape of weights and input/output tensors usually change due to such pruning.\nTo speedup this kind of pruning, there is no need to use sparse kernel, just replace the pruned layer with smaller one.\nSince the support of sparse kernels in community is limited,\nwe only support the speedup of coarse-grained pruning and leave the support of fine-grained pruning in future.\n\n## Design and Implementation\n\nTo speedup a model, the pruned layers should be replaced, either replaced with smaller layer for coarse-grained mask,\nor replaced with sparse kernel for fine-grained mask. Coarse-grained mask usually changes the shape of weights or input/output tensors,\nthus, we should do shape inference to check are there other unpruned layers should be replaced as well due to shape change.\nTherefore, in our design, there are two main steps: first, do shape inference to find out all the modules that should be replaced;\nsecond, replace the modules.\n\nThe first step requires topology (i.e., connections) of the model, we use ``jit.trace`` to obtain the model graph for PyTorch.\nThe new shape of module is auto-inference by NNI, the unchanged parts of outputs during forward and inputs during backward are prepared for reduct.\nFor each type of module, we should prepare a function for module replacement.\nThe module replacement function returns a newly created module which is smaller.\n\n## Usage\n"
"\n# Speedup Model with Mask\n\n## Introduction\n\nPruning algorithms usually use weight masks to simulate the real pruning. Masks can be used\nto check model performance of a specific pruning (or sparsity), but there is no real speedup.\nSince model speedup is the ultimate goal of model pruning, we try to provide a tool to users\nto convert a model to a smaller one based on user provided masks (the masks come from the\npruning algorithms).\n\nThere are two types of pruning. One is fine-grained pruning, it does not change the shape of weights,\nand input/output tensors. Sparse kernel is required to speedup a fine-grained pruned layer.\nThe other is coarse-grained pruning (e.g., channels), shape of weights and input/output tensors usually change due to such pruning.\nTo speedup this kind of pruning, there is no need to use sparse kernel, just replace the pruned layer with smaller one.\nSince the support of sparse kernels in community is limited,\nwe only support the speedup of coarse-grained pruning and leave the support of fine-grained pruning in future.\n\n## Design and Implementation\n\nTo speedup a model, the pruned layers should be replaced, either replaced with smaller layer for coarse-grained mask,\nor replaced with sparse kernel for fine-grained mask. Coarse-grained mask usually changes the shape of weights or input/output tensors,\nthus, we should do shape inference to check are there other unpruned layers should be replaced as well due to shape change.\nTherefore, in our design, there are two main steps: first, do shape inference to find out all the modules that should be replaced;\nsecond, replace the modules.\n\nThe first step requires topology (i.e., connections) of the model, we use ``jit.trace`` to obtain the model graph for PyTorch.\nThe new shape of module is auto-inference by NNI, the unchanged parts of outputs during forward and inputs during backward are prepared for reduct.\nFor each type of module, we should prepare a function for module replacement.\nThe module replacement function returns a newly created module which is smaller.\n\n## Usage\n"
]
},
{
...
...
@@ -76,7 +76,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"Speedup the model and show the model structure after speedup.\n\n"
"Speedup the model and show the model structure after speedup.\n\n"
]
},
{
...
...
@@ -94,7 +94,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"Roughly test the model after speed-up inference speed.\n\n"
"Roughly test the model after speedup inference speed.\n\n"
]
},
{
...
...
@@ -112,7 +112,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"For combining usage of ``Pruner`` masks generation with ``ModelSpeedup``,\nplease refer to `Pruning Quick Start <./pruning_quick_start_mnist.html>`__.\n\nNOTE: The current implementation supports PyTorch 1.3.1 or newer.\n\n## Limitations\n\nFor PyTorch we can only replace modules, if functions in ``forward`` should be replaced,\nour current implementation does not work. One workaround is make the function a PyTorch module.\n\nIf you want to speedup your own model which cannot supported by the current implementation,\nyou need implement the replace function for module replacement, welcome to contribute.\n\n## Speedup Results of Examples\n\nThe code of these experiments can be found :githublink:`here <examples/model_compress/pruning/speedup/model_speedup.py>`.\n\nThese result are tested on the `legacy pruning framework <../comporession/pruning_legacy>`__, new results will coming soon.\n\n### slim pruner example\n\non one V100 GPU,\ninput tensor: ``torch.randn(64, 3, 32, 32)``\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01197\n - 0.005107\n * - 2\n - 0.02019\n - 0.008769\n * - 4\n - 0.02733\n - 0.014809\n * - 8\n - 0.04310\n - 0.027441\n * - 16\n - 0.07731\n - 0.05008\n * - 32\n - 0.14464\n - 0.10027\n\n### fpgm pruner example\n\non cpu,\ninput tensor: ``torch.randn(64, 1, 28, 28)``\\ ,\ntoo large variance\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01383\n - 0.01839\n * - 2\n - 0.01167\n - 0.003558\n * - 4\n - 0.01636\n - 0.01088\n * - 40\n - 0.14412\n - 0.08268\n * - 40\n - 1.29385\n - 0.14408\n * - 40\n - 0.41035\n - 0.46162\n * - 400\n - 6.29020\n - 5.82143\n\n### l1filter pruner example\n\non one V100 GPU,\ninput tensor: ``torch.randn(64, 3, 32, 32)``\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01026\n - 0.003677\n * - 2\n - 0.01657\n - 0.008161\n * - 4\n - 0.02458\n - 0.020018\n * - 8\n - 0.03498\n - 0.025504\n * - 16\n - 0.06757\n - 0.047523\n * - 32\n - 0.10487\n - 0.086442\n\n### APoZ pruner example\n\non one V100 GPU,\ninput tensor: ``torch.randn(64, 3, 32, 32)``\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01389\n - 0.004208\n * - 2\n - 0.01628\n - 0.008310\n * - 4\n - 0.02521\n - 0.014008\n * - 8\n - 0.03386\n - 0.023923\n * - 16\n - 0.06042\n - 0.046183\n * - 32\n - 0.12421\n - 0.087113\n\n### SimulatedAnnealing pruner example\n\nIn this experiment, we use SimulatedAnnealing pruner to prune the resnet18 on the cifar10 dataset.\nWe measure the latencies and accuracies of the pruned model under different sparsity ratios, as shown in the following figure.\nThe latency is measured on one V100 GPU and the input tensor is ``torch.randn(128, 3, 32, 32)``.\n\n<img src=\"file://../../img/SA_latency_accuracy.png\">\n\n"
"For combining usage of ``Pruner`` masks generation with ``ModelSpeedup``,\nplease refer to `Pruning Quick Start <./pruning_quick_start_mnist.html>`__.\n\nNOTE: The current implementation supports PyTorch 1.3.1 or newer.\n\n## Limitations\n\nFor PyTorch we can only replace modules, if functions in ``forward`` should be replaced,\nour current implementation does not work. One workaround is make the function a PyTorch module.\n\nIf you want to speedup your own model which cannot supported by the current implementation,\nyou need implement the replace function for module replacement, welcome to contribute.\n\n## Speedup Results of Examples\n\nThe code of these experiments can be found :githublink:`here <examples/model_compress/pruning/speedup/model_speedup.py>`.\n\nThese result are tested on the `legacy pruning framework <../comporession/pruning_legacy>`__, new results will coming soon.\n\n### slim pruner example\n\non one V100 GPU,\ninput tensor: ``torch.randn(64, 3, 32, 32)``\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01197\n - 0.005107\n * - 2\n - 0.02019\n - 0.008769\n * - 4\n - 0.02733\n - 0.014809\n * - 8\n - 0.04310\n - 0.027441\n * - 16\n - 0.07731\n - 0.05008\n * - 32\n - 0.14464\n - 0.10027\n\n### fpgm pruner example\n\non cpu,\ninput tensor: ``torch.randn(64, 1, 28, 28)``\\ ,\ntoo large variance\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01383\n - 0.01839\n * - 2\n - 0.01167\n - 0.003558\n * - 4\n - 0.01636\n - 0.01088\n * - 40\n - 0.14412\n - 0.08268\n * - 40\n - 1.29385\n - 0.14408\n * - 40\n - 0.41035\n - 0.46162\n * - 400\n - 6.29020\n - 5.82143\n\n### l1filter pruner example\n\non one V100 GPU,\ninput tensor: ``torch.randn(64, 3, 32, 32)``\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01026\n - 0.003677\n * - 2\n - 0.01657\n - 0.008161\n * - 4\n - 0.02458\n - 0.020018\n * - 8\n - 0.03498\n - 0.025504\n * - 16\n - 0.06757\n - 0.047523\n * - 32\n - 0.10487\n - 0.086442\n\n### APoZ pruner example\n\non one V100 GPU,\ninput tensor: ``torch.randn(64, 3, 32, 32)``\n\n.. list-table::\n :header-rows: 1\n :widths: auto\n\n * - Times\n - Mask Latency\n - Speedup Latency\n * - 1\n - 0.01389\n - 0.004208\n * - 2\n - 0.01628\n - 0.008310\n * - 4\n - 0.02521\n - 0.014008\n * - 8\n - 0.03386\n - 0.023923\n * - 16\n - 0.06042\n - 0.046183\n * - 32\n - 0.12421\n - 0.087113\n\n### SimulatedAnnealing pruner example\n\nIn this experiment, we use SimulatedAnnealing pruner to prune the resnet18 on the cifar10 dataset.\nWe measure the latencies and accuracies of the pruned model under different sparsity ratios, as shown in the following figure.\nThe latency is measured on one V100 GPU and the input tensor is ``torch.randn(128, 3, 32, 32)``.\n\n<img src=\"file://../../img/SA_latency_accuracy.png\">\n\n"
"\n# SpeedUp Model with Calibration Config\n\n\n## Introduction\n\nDeep learning network has been computational intensive and memory intensive \nwhich increases the difficulty of deploying deep neural network model. Quantization is a \nfundamental technology which is widely used to reduce memory footprint and speedup inference \nprocess. Many frameworks begin to support quantization, but few of them support mixed precision \nquantization and get real speedup. Frameworks like `HAQ: Hardware-Aware Automated Quantization with Mixed Precision <https://arxiv.org/pdf/1811.08886.pdf>`__\\, only support simulated mixed precision quantization which will \nnot speedup the inference process. To get real speedup of mixed precision quantization and \nhelp people get the real feedback from hardware, we design a general framework with simple interface to allow NNI quantization algorithms to connect different \nDL model optimization backends (e.g., TensorRT, NNFusion), which gives users an end-to-end experience that after quantizing their model \nwith quantization algorithms, the quantized model can be directly speeded up with the connected optimization backend. NNI connects \nTensorRT at this stage, and will support more backends in the future.\n\n\n## Design and Implementation\n\nTo support speeding up mixed precision quantization, we divide framework into two part, frontend and backend. \nFrontend could be popular training frameworks such as PyTorch, TensorFlow etc. Backend could be inference \nframework for different hardwares, such as TensorRT. At present, we support PyTorch as frontend and \nTensorRT as backend. To convert PyTorch model to TensorRT engine, we leverage onnx as intermediate graph \nrepresentation. In this way, we convert PyTorch model to onnx model, then TensorRT parse onnx \nmodel to generate inference engine. \n\n\nQuantization aware training combines NNI quantization algorithm 'QAT' and NNI quantization speedup tool.\nUsers should set config to train quantized model using QAT algorithm(please refer to `NNI Quantization Algorithms <https://nni.readthedocs.io/en/stable/Compression/Quantizer.html>`__\\ ).\nAfter quantization aware training, users can get new config with calibration parameters and model with quantized weight. By passing new config and model to quantization speedup tool, users can get real mixed precision speedup engine to do inference.\n\n\nAfter getting mixed precision engine, users can do inference with input data.\n\n\nNote\n\n\n* Recommend using \"cpu\"(host) as data device(for both inference data and calibration data) since data should be on host initially and it will be transposed to device before inference. If data type is not \"cpu\"(host), this tool will transpose it to \"cpu\" which may increases unnecessary overhead.\n* User can also do post-training quantization leveraging TensorRT directly(need to provide calibration dataset).\n* Not all op types are supported right now. At present, NNI supports Conv, Linear, Relu and MaxPool. More op types will be supported in the following release.\n\n\n## Prerequisite\nCUDA version >= 11.0\n\nTensorRT version >= 7.2\n\nNote\n\n* If you haven't installed TensorRT before or use the old version, please refer to `TensorRT Installation Guide <https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html>`__\\ \n\n## Usage\n"
"\n# SpeedUp Model with Calibration Config\n\n\n## Introduction\n\nDeep learning network has been computational intensive and memory intensive \nwhich increases the difficulty of deploying deep neural network model. Quantization is a \nfundamental technology which is widely used to reduce memory footprint and speedup inference \nprocess. Many frameworks begin to support quantization, but few of them support mixed precision \nquantization and get real speedup. Frameworks like `HAQ: Hardware-Aware Automated Quantization with Mixed Precision <https://arxiv.org/pdf/1811.08886.pdf>`__\\, only support simulated mixed precision quantization which will \nnot speedup the inference process. To get real speedup of mixed precision quantization and \nhelp people get the real feedback from hardware, we design a general framework with simple interface to allow NNI quantization algorithms to connect different \nDL model optimization backends (e.g., TensorRT, NNFusion), which gives users an end-to-end experience that after quantizing their model \nwith quantization algorithms, the quantized model can be directly speeded up with the connected optimization backend. NNI connects \nTensorRT at this stage, and will support more backends in the future.\n\n\n## Design and Implementation\n\nTo support speeding up mixed precision quantization, we divide framework into two part, frontend and backend. \nFrontend could be popular training frameworks such as PyTorch, TensorFlow etc. Backend could be inference \nframework for different hardwares, such as TensorRT. At present, we support PyTorch as frontend and \nTensorRT as backend. To convert PyTorch model to TensorRT engine, we leverage onnx as intermediate graph \nrepresentation. In this way, we convert PyTorch model to onnx model, then TensorRT parse onnx \nmodel to generate inference engine. \n\n\nQuantization aware training combines NNI quantization algorithm 'QAT' and NNI quantization speedup tool.\nUsers should set config to train quantized model using QAT algorithm(please refer to `NNI Quantization Algorithms <https://nni.readthedocs.io/en/stable/Compression/Quantizer.html>`__\\ ).\nAfter quantization aware training, users can get new config with calibration parameters and model with quantized weight. By passing new config and model to quantization speedup tool, users can get real mixed precision speedup engine to do inference.\n\n\nAfter getting mixed precision engine, users can do inference with input data.\n\n\nNote\n\n\n* Recommend using \"cpu\"(host) as data device(for both inference data and calibration data) since data should be on host initially and it will be transposed to device before inference. If data type is not \"cpu\"(host), this tool will transpose it to \"cpu\" which may increases unnecessary overhead.\n* User can also do post-training quantization leveraging TensorRT directly(need to provide calibration dataset).\n* Not all op types are supported right now. At present, NNI supports Conv, Linear, Relu and MaxPool. More op types will be supported in the following release.\n\n\n## Prerequisite\nCUDA version >= 11.0\n\nTensorRT version >= 7.2\n\nNote\n\n* If you haven't installed TensorRT before or use the old version, please refer to `TensorRT Installation Guide <https://docs.nvidia.com/deeplearning/tensorrt/install-guide/index.html>`__\\ \n\n## Usage\n"
]
},
{
...
...
@@ -69,7 +69,7 @@
"cell_type": "markdown",
"metadata": {},
"source": [
"build tensorRT engine to make a real speedup\n\n"
"build tensorRT engine to make a real speedup\n\n"
Deep learning network has been computational intensive and memory intensive
which increases the difficulty of deploying deep neural network model. Quantization is a
fundamental technology which is widely used to reduce memory footprint and speedup inference
fundamental technology which is widely used to reduce memory footprint and speedup inference
process. Many frameworks begin to support quantization, but few of them support mixed precision
quantization and get real speedup. Frameworks like `HAQ: Hardware-Aware Automated Quantization with Mixed Precision <https://arxiv.org/pdf/1811.08886.pdf>`__\, only support simulated mixed precision quantization which will
not speedup the inference process. To get real speedup of mixed precision quantization and
not speedup the inference process. To get real speedup of mixed precision quantization and
help people get the real feedback from hardware, we design a general framework with simple interface to allow NNI quantization algorithms to connect different
DL model optimization backends (e.g., TensorRT, NNFusion), which gives users an end-to-end experience that after quantizing their model
with quantization algorithms, the quantized model can be directly speeded up with the connected optimization backend. NNI connects
