Tensorflow中float32模型强制转为float16半浮点模型
在Tensorflow框架训练完成后，部署模型时希望对模型进⾏压缩。⼀种⽅案是前⾯⽂字介绍的⽅法。另⼀种⽅法是半浮点量化，今天我们主要介绍如何通过修改Tensorflow的pb⽂件中的计算节点和常量（const），将float32数据类型的模型⼤⼩压缩减半为float16数据类型的模型。
最近看到⼀个巨⽜的⼈⼯智能教程，分享⼀下给⼤家。教程不仅是零基础，通俗易懂，⽽且⾮常风趣幽默，像看⼩说⼀样！觉得太⽜了，所以分享给⼤家。平时碎⽚时间可以当⼩说看，。
1 加载pb模型
封装函数，加载pb模型：
def load_graph(model_path):
graph = tf.Graph()
with graph.as_default():
graph_def = tf.GraphDef()
if dswith("pb"):
with open(model_path, "rb") as f:
graph_def.ad())
else:
with open(model_path, "r") as pf:
text_format.ad(), graph_def)
tf.import_graph_def(graph_def, name="")
sess = tf.Session(graph=graph)
_operations()
for op in ops:
print(op.name)
return sess
2 重写BatchNorm
由于BatchNorm对精度⽐较敏感，需要保持float32类型，因此BatchNorm需要特殊处理。
#⽤FusedBatchNormV2替换FusedBatchNorm，以保证反向梯度下降计算时使⽤的是float
def rewrite_batch_norm_node_v2(node, graph_def, target_type='fp16'):
if target_type == 'fp16':
dtype = types_pb2.DT_HALF
elif target_type == 'fp64':
dtype = types_pb2.DT_DOUBLE
else:
dtype = types_pb2.DT_FLOAT
new_node = de.add()
new_node.op = "FusedBatchNormV2"
new_node.name = node.name
new_d(node.input)
new_node.attr["U"].CopyFrom(attr_value_pb2.AttrValue(type=types_pb2.DT_FLOAT))
for attr in list(node.attr.keys()):
if attr == "T":
node.attr[attr].type = dtype
new_node.attr[attr].CopyFrom(node.attr[attr])
print("rewrite fused_batch_norm done!")
3 Graph转换
重新构造graph，参数从原始pb的graph中拷贝，并转为float16
def convert_graph_to_fp16(model_path, save_path, name, as_text=False, target_type='fp16', input_name=None, output_names=None): #⽣成新的图数据类型
if target_type == 'fp16':
dtype = types_pb2.DT_HALF
elif target_type == 'fp64':
dtype = types_pb2.DT_DOUBLE
else:
dtype = types_pb2.DT_FLOAT
#加载需要转换的模型
source_sess = load_graph(model_path)
source_graph_def = aph.as_graph_def()
#创建新的模图对象
target_graph_def = graph_pb2.GraphDef()
target_graph_def.versions.CopyFrom(source_graph_def.versions)
#对加载的模型遍历计算节点
for node in source_de:
# 对FusedBatchNorm计算节点替换为FusedBatchNormV2
if node.op == "FusedBatchNorm":
rewrite_batch_norm_node_v2(node, target_graph_def, target_type=target_type)
continue
# 复制计算节点
new_node = target_de.add()
new_node.op = node.op
new_node.name = node.name
new_d(node.input)
#对attrs属性进⾏复制，attrs属性主要关注
attrs = list(node.attr.keys())
# BatchNorm属性保持不变
if ("BatchNorm" in node.name) or ('batch_normalization' in node.name):
for attr in attrs:
new_node.attr[attr].CopyFrom(node.attr[attr])
continue
# 除了BatchNorm以外其他计算节点的属性单独
for attr in attrs:
# 对指定的计算节点保持不变
if node.name in keep_fp32_node_name:
new_node.attr[attr].CopyFrom(node.attr[attr])
continue
#将Float类型修改为设置的⽬标类型
if node.attr[attr].type == types_pb2.DT_FLOAT:
# modify node dtype
node.attr[attr].type = dtype
#重点关注value，weights都是保存在value属性中
if attr == "value":
tensor = node.attr[attr].tensor
if tensor.dtype == types_pb2.DT_FLOAT:
# if float_val exists
if tensor.float_val:
float_val = tf.make_ndarray(node.attr[attr].tensor)
new_node.attr[attr].tensor.CopyFrom(tf.make_tensor_proto(float_val, dtype=dtype))
continue
# if tensor content exists
sor_content:
tensor_shape = [x.size for x sor_shape.dim]
tensor_weights = tf.make_ndarray(tensor)
# reshape tensor
tensor_weights = np.reshape(tensor_weights, tensor_shape)
tensor_proto = tf.make_tensor_proto(tensor_weights, dtype=dtype)
new_node.attr[attr].tensor.CopyFrom(tensor_proto)
continue
new_node.attr[attr].CopyFrom(node.attr[attr])
# transform graph
# transform graph
if output_names:
if not input_name:
input_name = []
transforms = ["strip_unused_nodes"]
target_graph_def = TransformGraph(target_graph_def, input_name, output_names, transforms)
# write graph_def to model
tf.io.write_graph(target_graph_def, logdir=save_path, name=name, as_text=as_text)
print("Converting done ...")
4 完整的代码
import tensorflow as tf
framework import types_pb2, graph_pb2, attr_value_pb2
ls.graph_transforms import TransformGraph
from google.protobuf import text_format
import numpy as np
# object detection api input and output nodes
input_name = "input_tf"
output_names = ["output:0"]
keep_fp32_node_name = []
def load_graph(model_path):
graph = tf.Graph()
with graph.as_default():
graph_def = tf.GraphDef()
if dswith("pb"):
with open(model_path, "rb") as f:
graph_def.ad())
else:
with open(model_path, "r") as pf:
text_format.ad(), graph_def)
tf.import_graph_def(graph_def, name="")
sess = tf.Session(graph=graph)
_operations()
for op in ops:
print(op.name)
return sess
#⽤FusedBatchNormV2替换FusedBatchNorm，以保证反向梯度下降计算时使⽤的是float
def rewrite_batch_norm_node_v2(node, graph_def, target_type='fp16'):
if target_type == 'fp16':
dtype = types_pb2.DT_HALF
elif target_type == 'fp64':
dtype = types_pb2.DT_DOUBLE
else:
dtype = types_pb2.DT_FLOAT
new_node = de.add()
new_node.op = "FusedBatchNormV2"
new_node.name = node.name
new_d(node.input)
new_node.attr["U"].CopyFrom(attr_value_pb2.AttrValue(type=types_pb2.DT_FLOAT))
for attr in list(node.attr.keys()):
if attr == "T":
node.attr[attr].type = dtype
new_node.attr[attr].CopyFrom(node.attr[attr])
print("rewrite fused_batch_norm done!")
def convert_graph_to_fp16(model_path, save_path, name, as_text=False, target_type='fp16', input_name=None, output_names=None): #⽣成新的图数据类型
if target_type == 'fp16':
dtype = types_pb2.DT_HALF
elif target_type == 'fp64':
dtype = types_pb2.DT_DOUBLE
dtype = types_pb2.DT_DOUBLE
else:
dtype = types_pb2.DT_FLOAT
#加载需要转换的模型
source_sess = load_graph(model_path)
source_graph_def = aph.as_graph_def()
#创建新的模图对象
target_graph_def = graph_pb2.GraphDef()
target_graph_def.versions.CopyFrom(source_graph_def.versions)
#对加载的模型遍历计算节点
for node in source_de:
# 对FusedBatchNorm计算节点替换为FusedBatchNormV2
if node.op == "FusedBatchNorm":
rewrite_batch_norm_node_v2(node, target_graph_def, target_type=target_type)
continue
# 复制计算节点
new_node = target_de.add()
new_node.op = node.op
new_node.name = node.name
new_d(node.input)
#对attrs属性进⾏复制，attrs属性主要关注
attrs = list(node.attr.keys())
# BatchNorm属性保持不变
if ("BatchNorm" in node.name) or ('batch_normalization' in node.name):
for attr in attrs:
new_node.attr[attr].CopyFrom(node.attr[attr])
continue
# 除了BatchNorm以外其他计算节点的属性单独
for attr in attrs:
# 对指定的计算节点保持不变
if node.name in keep_fp32_node_name:
new_node.attr[attr].CopyFrom(node.attr[attr])
continue
#将Float类型修改为设置的⽬标类型
if node.attr[attr].type == types_pb2.DT_FLOAT:
# modify node dtype
node.attr[attr].type = dtype
transform和convert的区别#重点关注value，weights都是保存在value属性中
if attr == "value":
tensor = node.attr[attr].tensor
if tensor.dtype == types_pb2.DT_FLOAT:
# if float_val exists
if tensor.float_val:
float_val = tf.make_ndarray(node.attr[attr].tensor)
new_node.attr[attr].tensor.CopyFrom(tf.make_tensor_proto(float_val, dtype=dtype)) continue
# if tensor content exists
sor_content:
tensor_shape = [x.size for x sor_shape.dim]
tensor_weights = tf.make_ndarray(tensor)
# reshape tensor
tensor_weights = np.reshape(tensor_weights, tensor_shape)
tensor_proto = tf.make_tensor_proto(tensor_weights, dtype=dtype)
new_node.attr[attr].tensor.CopyFrom(tensor_proto)
continue
new_node.attr[attr].CopyFrom(node.attr[attr])
# transform graph
if output_names:
if not input_name:
input_name = []
transforms = ["strip_unused_nodes"]
target_graph_def = TransformGraph(target_graph_def, input_name, output_names, transforms) # write graph_def to model
# write graph_def to model
tf.io.write_graph(target_graph_def, logdir=save_path, name=name, as_text=as_text)
print("Converting done ...")
save_path = "test"
name = "output_fp16.pb"
model_path="test.pb"
as_text = False
target_type = 'fp16'
convert_graph_to_fp16(model_path, save_path, name, as_text=as_text, target_type=target_type, input_name=input_name, output_names=output_names)
# 测试⼀下转换后的模型是否能够加载
sess = load_graph(save_path+"/"+name)
版权声明：本站内容均来自互联网，仅供演示用，请勿用于商业和其他非法用途。如果侵犯了您的权益请与我们联系QQ:729038198，我们将在24小时内删除。
发表评论