Help protect the Great Barrier Reef with TensorFlow on Kaggle Join Challenge

训练后 float16 量化

在 TensorFlow.org 上查看 在 Google Colab 中运行 在 GitHub 上查看源代码 下载笔记本

概述

现在,TensorFlow Lite 支持在模型从 TensorFlow 转换到 TensorFlow Lite FlatBuffer 格式期间将权重转换为 16 位浮点值。这样可以将模型的大小缩减至原来的二分之一。某些硬件(如 GPU)可以在这种精度降低的算术中以原生方式计算,从而实现比传统浮点执行更快的速度。可以将 Tensorflow Lite GPU 委托配置为以这种方式运行。但是,转换为 float16 权重的模型仍可在 CPU 上运行而无需其他修改:float16 权重会在首次推断前上采样为 float32。这样可以在对延迟和准确率造成最小影响的情况下显著缩减模型大小。

在本教程中,您将从头开始训练一个 MNIST 模型,并在 TensorFlow 中检查其准确率,然后使用 float16 量化将此模型转换为 Tensorflow Lite FlatBuffer 格式。最后,检查转换后模型的准确率,并将其与原始 float32 模型进行比较。

构建 MNIST 模型

设置

import logging
logging.getLogger("tensorflow").setLevel(logging.DEBUG)

import tensorflow as tf
from tensorflow import keras
import numpy as np
import pathlib
tf.float16
tf.float16

训练并导出模型

# Load MNIST dataset
mnist = keras.datasets.mnist
(train_images, train_labels), (test_images, test_labels) = mnist.load_data()

# Normalize the input image so that each pixel value is between 0 to 1.
train_images = train_images / 255.0
test_images = test_images / 255.0

# Define the model architecture
model = keras.Sequential([
  keras.layers.InputLayer(input_shape=(28, 28)),
  keras.layers.Reshape(target_shape=(28, 28, 1)),
  keras.layers.Conv2D(filters=12, kernel_size=(3, 3), activation=tf.nn.relu),
  keras.layers.MaxPooling2D(pool_size=(2, 2)),
  keras.layers.Flatten(),
  keras.layers.Dense(10)
])

# Train the digit classification model
model.compile(optimizer='adam',
              loss=keras.losses.SparseCategoricalCrossentropy(from_logits=True),
              metrics=['accuracy'])
model.fit(
  train_images,
  train_labels,
  epochs=1,
  validation_data=(test_images, test_labels)
)
2021-08-13 21:08:17.716181: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:17.724210: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:17.725104: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:17.726698: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA
To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.
2021-08-13 21:08:17.727282: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:17.728211: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:17.729036: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:18.306135: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:18.307039: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:18.307865: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:18.308683: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 14648 MB memory:  -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:05.0, compute capability: 7.0
2021-08-13 21:08:19.176727: I tensorflow/compiler/mlir/mlir_graph_optimization_pass.cc:185] None of the MLIR Optimization Passes are enabled (registered 2)
2021-08-13 21:08:19.904497: I tensorflow/stream_executor/cuda/cuda_dnn.cc:369] Loaded cuDNN version 8100
2021-08-13 21:08:20.426817: I tensorflow/core/platform/default/subprocess.cc:304] Start cannot spawn child process: No such file or directory
1875/1875 [==============================] - 6s 2ms/step - loss: 0.2768 - accuracy: 0.9226 - val_loss: 0.1293 - val_accuracy: 0.9643
<keras.callbacks.History at 0x7fdff8631cd0>

在此示例中,您只对模型进行了一个周期的训练,因此只训练到约 96% 的准确率。

转换为 TensorFlow Lite 模型

现在,您可以使用 Python TFLiteConverter 将训练的模型转换为 TensorFlow Lite 模型。

现在使用 TFLiteConverter 加载模型:

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
2021-08-13 21:08:24.926050: W tensorflow/python/util/util.cc:348] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.
INFO:tensorflow:Assets written to: /tmp/tmphddetpcb/assets
2021-08-13 21:08:25.296210: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.296548: I tensorflow/core/grappler/devices.cc:66] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 1
2021-08-13 21:08:25.296644: I tensorflow/core/grappler/clusters/single_machine.cc:357] Starting new session
2021-08-13 21:08:25.297011: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.297351: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.297627: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.297948: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.298221: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.298461: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 14648 MB memory:  -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:05.0, compute capability: 7.0
2021-08-13 21:08:25.299791: I tensorflow/core/grappler/optimizers/meta_optimizer.cc:1137] Optimization results for grappler item: graph_to_optimize
  function_optimizer: function_optimizer did nothing. time = 0.007ms.
  function_optimizer: function_optimizer did nothing. time = 0.001ms.

2021-08-13 21:08:25.330215: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:351] Ignored output_format.
2021-08-13 21:08:25.330246: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:354] Ignored drop_control_dependency.
2021-08-13 21:08:25.333269: I tensorflow/compiler/mlir/tensorflow/utils/dump_mlir_util.cc:210] disabling MLIR crash reproducer, set env var `MLIR_CRASH_REPRODUCER_DIRECTORY` to enable.

将其写入 .tflite 文件:

tflite_models_dir = pathlib.Path("/tmp/mnist_tflite_models/")
tflite_models_dir.mkdir(exist_ok=True, parents=True)
tflite_model_file = tflite_models_dir/"mnist_model.tflite"
tflite_model_file.write_bytes(tflite_model)
84500

要改为在导出时将模型量化为 float16,首先将 optimizations 标记设置为使用默认优化。然后将 float16 指定为目标平台支持的类型:

converter.optimizations = [tf.lite.Optimize.DEFAULT]
converter.target_spec.supported_types = [tf.float16]

最后,像往常一样转换模型。请注意,为了方便调用,转换后的模型默认仍将使用浮点输入和输出。

tflite_fp16_model = converter.convert()
tflite_model_fp16_file = tflite_models_dir/"mnist_model_quant_f16.tflite"
tflite_model_fp16_file.write_bytes(tflite_fp16_model)
INFO:tensorflow:Assets written to: /tmp/tmp92w6zo2k/assets
INFO:tensorflow:Assets written to: /tmp/tmp92w6zo2k/assets
2021-08-13 21:08:25.873772: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.874101: I tensorflow/core/grappler/devices.cc:66] Number of eligible GPUs (core count >= 8, compute capability >= 0.0): 1
2021-08-13 21:08:25.874190: I tensorflow/core/grappler/clusters/single_machine.cc:357] Starting new session
2021-08-13 21:08:25.874552: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.874871: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.875126: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.875465: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.875733: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero
2021-08-13 21:08:25.875970: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 14648 MB memory:  -> device: 0, name: Tesla V100-SXM2-16GB, pci bus id: 0000:00:05.0, compute capability: 7.0
2021-08-13 21:08:25.877336: I tensorflow/core/grappler/optimizers/meta_optimizer.cc:1137] Optimization results for grappler item: graph_to_optimize
  function_optimizer: function_optimizer did nothing. time = 0.008ms.
  function_optimizer: function_optimizer did nothing. time = 0.001ms.

2021-08-13 21:08:25.909015: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:351] Ignored output_format.
2021-08-13 21:08:25.909051: W tensorflow/compiler/mlir/lite/python/tf_tfl_flatbuffer_helpers.cc:354] Ignored drop_control_dependency.
44432

请注意,生成文件的大小约为 1/2

ls -lh {tflite_models_dir}
total 128K
-rw-rw-r-- 1 kbuilder kbuilder 83K Aug 13 21:08 mnist_model.tflite
-rw-rw-r-- 1 kbuilder kbuilder 44K Aug 13 21:08 mnist_model_quant_f16.tflite

运行 TensorFlow Lite 模型

使用 Python TensorFlow Lite 解释器运行 TensorFlow Lite 模型。

将模型加载到解释器中

interpreter = tf.lite.Interpreter(model_path=str(tflite_model_file))
interpreter.allocate_tensors()
interpreter_fp16 = tf.lite.Interpreter(model_path=str(tflite_model_fp16_file))
interpreter_fp16.allocate_tensors()

在单个图像上测试模型

test_image = np.expand_dims(test_images[0], axis=0).astype(np.float32)

input_index = interpreter.get_input_details()[0]["index"]
output_index = interpreter.get_output_details()[0]["index"]

interpreter.set_tensor(input_index, test_image)
interpreter.invoke()
predictions = interpreter.get_tensor(output_index)
import matplotlib.pylab as plt

plt.imshow(test_images[0])
template = "True:{true}, predicted:{predict}"
_ = plt.title(template.format(true= str(test_labels[0]),
                              predict=str(np.argmax(predictions[0]))))
plt.grid(False)

png

test_image = np.expand_dims(test_images[0], axis=0).astype(np.float32)

input_index = interpreter_fp16.get_input_details()[0]["index"]
output_index = interpreter_fp16.get_output_details()[0]["index"]

interpreter_fp16.set_tensor(input_index, test_image)
interpreter_fp16.invoke()
predictions = interpreter_fp16.get_tensor(output_index)
plt.imshow(test_images[0])
template = "True:{true}, predicted:{predict}"
_ = plt.title(template.format(true= str(test_labels[0]),
                              predict=str(np.argmax(predictions[0]))))
plt.grid(False)

png

评估模型

# A helper function to evaluate the TF Lite model using "test" dataset.
def evaluate_model(interpreter):
  input_index = interpreter.get_input_details()[0]["index"]
  output_index = interpreter.get_output_details()[0]["index"]

  # Run predictions on every image in the "test" dataset.
  prediction_digits = []
  for test_image in test_images:
    # Pre-processing: add batch dimension and convert to float32 to match with
    # the model's input data format.
    test_image = np.expand_dims(test_image, axis=0).astype(np.float32)
    interpreter.set_tensor(input_index, test_image)

    # Run inference.
    interpreter.invoke()

    # Post-processing: remove batch dimension and find the digit with highest
    # probability.
    output = interpreter.tensor(output_index)
    digit = np.argmax(output()[0])
    prediction_digits.append(digit)

  # Compare prediction results with ground truth labels to calculate accuracy.
  accurate_count = 0
  for index in range(len(prediction_digits)):
    if prediction_digits[index] == test_labels[index]:
      accurate_count += 1
  accuracy = accurate_count * 1.0 / len(prediction_digits)

  return accuracy
print(evaluate_model(interpreter))
0.9643

在 float16 量化模型上重复评估,以获得如下结果:

# NOTE: Colab runs on server CPUs. At the time of writing this, TensorFlow Lite
# doesn't have super optimized server CPU kernels. For this reason this may be
# slower than the above float interpreter. But for mobile CPUs, considerable
# speedup can be observed.
print(evaluate_model(interpreter_fp16))
0.9644

在此示例中,您已将模型量化为 float16,但准确率没有任何差别。

您还可以在 GPU 上评估 fp16 量化模型。要使用降低的精度值执行所有算术,请确保在您的应用中创建 TfLiteGPUDelegateOptions 结构,并将 precision_loss_allowed 设置为 1,如下所示:

//Prepare GPU delegate. const TfLiteGpuDelegateOptions options = {   .metadata = NULL,   .compile_options = {     .precision_loss_allowed = 1,  // FP16     .preferred_gl_object_type = TFLITE_GL_OBJECT_TYPE_FASTEST,     .dynamic_batch_enabled = 0,   // Not fully functional yet   }, };

有关 TFLite GPU 委托以及如何在您的应用中进行使用的详细文档,请参阅此处