我正在尝试通过TF部署更大的Google AI Platform模型。
训练并保存为1.12。
最初，我的模型格式为
96B May 16 15:45 assets
142K May 16 15:45 saved_model.pb
160B May 17 12:14 variables
./assets:
8.1M May 16 15:45 tokens.txt
./variables:
8B May 16 15:45 variables.data-00000-of-00002
951M May 16 15:45 variables.data-00001-of-00002
559B May 16 15:45 variables.index
这太大（限制为256 Mb）。因此，然后我使用下面的脚本执行了一些痛苦的过程，冻结，优化和重新保存。
"""
A set of functions that freeze,optimise and re-save and TF model.
Purely to decrease the model size for Google Cloud.
Model trained with: TF 1.12.0
This script can only be run with: TF 1.14.1-dev20190516 (a nightly)
mkdir -p frozen/1
mkdir -p optimised/1
mkdir -p final_model/1
Python reduce_model_size.py
"""
import os
import numpy as np
from datetime import datetime
import sys
import tensorflow as tf
from tensorflow.python.saved_model import tag_constants
from tensorflow.python.tools import freeze_graph
from tensorflow.python import ops
from tensorflow.tools.graph_transforms import TransformGraph
def get_graph_def_from_saved_model(saved_model_dir):
with tf.Session() as session:
meta_graph_def = tf.saved_model.loader.load(
session,
tags=[tf.saved_model.tag_constants.SERVING],
export_dir=saved_model_dir
)
return meta_graph_def.graph_def
def describe_graph(graph_def, show_nodes=False):
print('Input Feature Nodes: {}'.format(
[node.name for node in graph_def.node if node.op=='Placeholder']))
print('')
print('Unused Nodes: {}'.format(
[node.name for node in graph_def.node if 'unused' in node.name]))
print('')
print('Output Nodes: {}'.format(
[node.name for node in graph_def.node if (
'predictions' in node.name or 'softmax' in node.name)]))
print('')
print('Quantization Nodes: {}'.format(
[node.name for node in graph_def.node if 'quant' in node.name]))
print('')
print('Constant Count: {}'.format(
len([node for node in graph_def.node if node.op=='Const'])))
print('')
print('Variable Count: {}'.format(
len([node for node in graph_def.node if 'Variable' in node.op])))
print('')
print('Identity Count: {}'.format(
len([node for node in graph_def.node if node.op=='Identity'])))
print('', 'Total nodes: {}'.format(len(graph_def.node)), '')
if show_nodes==True:
for node in graph_def.node:
print('Op:{} - Name: {}'.format(node.op, node.name))
def get_size(model_dir, model_file='saved_model.pb'):
model_file_path = os.path.join(model_dir, model_file)
print(model_file_path, '')
pb_size = os.path.getsize(model_file_path)
variables_size = 0
if os.path.exists(
os.path.join(model_dir,'variables/variables.data-00000-of-00002')):
variables_size = os.path.getsize(os.path.join(
model_dir,'variables/variables.data-00001-of-00002'))
variables_size += os.path.getsize(os.path.join(
model_dir,'variables/variables.index'))
print('Model size: {} MB'.format(round(pb_size/(1.049e+6),3)))
print('Variables size: {} MB'.format(round( variables_size/(1.049e+6),3)))
print('Total Size: {} MB'.format(round((pb_size + variables_size)/(1.049e+6),3)))
def freeze_model(saved_model_dir, output_node_names, output_filename):
output_graph_filename = output_filename
initializer_nodes = ''
freeze_graph.freeze_graph(
input_saved_model_dir=saved_model_dir,
output_graph=output_graph_filename,
saved_model_tags = tf.saved_model.tag_constants.SERVING,
output_node_names=output_node_names,
initializer_nodes=initializer_nodes,
input_graph=None,
input_saver=False,
input_binary=False,
input_checkpoint=None,
restore_op_name=False,
filename_tensor_name=None,
clear_devices=True,
input_meta_graph=False,
)
print('graph frozen!')
def get_graph_def_from_file(graph_filepath):
with ops.Graph().as_default():
with tf.gfile.GFile(graph_filepath, 'rb') as f:
graph_def = tf.GraphDef()
graph_def.ParseFromString(f.read())
return graph_def
def optimize_graph(model_dir, graph_filename,output_dir,optimised_file, transforms, output_node):
input_names = []
output_names = [output_node]
if graph_filename is None:
graph_def = get_graph_def_from_saved_model(model_dir)
else:
graph_def = get_graph_def_from_file(os.path.join(model_dir, graph_filename))
optimized_graph_def = TransformGraph(
graph_def,
input_names,
output_names,
transforms)
tf.train.write_graph(optimized_graph_def,
logdir=output_dir,
as_text=False,
name=optimised_file)
print('Graph optimized!')
def convert_graph_def_to_saved_model(export_dir, graph_filepath):
if tf.gfile.Exists(export_dir):
tf.gfile.DeleteRecursively(export_dir)
graph_def = get_graph_def_from_file(graph_filepath)
legacy_init_op = tf.group(tf.tables_initializer(), name='legacy_init_op')
with tf.Session(graph=tf.Graph()) as session:
tf.import_graph_def(graph_def, name='')
tf.saved_model.simple_save(
session,
export_dir,
inputs={
node.name: session.graph.get_tensor_by_name(
'{}:0'.format(node.name))
for node in graph_def.node if node.op=='Placeholder'},
outputs={'class_ids': session.graph.get_tensor_by_name(
'dnn/head/predictions/class_ids:0')},
legacy_init_op=legacy_init_op
)
print('Optimized graph converted to SavedModel!')
saved_model_dir="sentiment/1/"
print("*"*100)
print("Pre-frozen")
graph = get_graph_def_from_saved_model(saved_model_dir)
describe_graph(graph)
get_size(saved_model_dir,"saved_model.pb")
print("*"*100)
print("Frozen model: ")
frozen_dir = "frozen/1/"
frozen_file="saved_model.pb"
freeze_model(saved_model_dir,"dnn/head/predictions/class_ids",frozen_dir+frozen_file)
describe_graph(get_graph_def_from_file(frozen_dir+frozen_file))
get_size(frozen_dir,frozen_file)
print("*"*100)
transforms = [
'remove_nodes(op=Identity)',
'merge_duplicate_nodes',
'strip_unused_nodes',
'fold_constants(ignore_errors=true)',
'fold_batch_norms',
'quantize_nodes',
'quantize_weights',
'remove_attribute(attribute_name=_class)' # Solves the expects to be colocated with unknown node error
]
optimised_dir = "optimised/1/"
optimised_file="saved_model.pb"
optimize_graph(frozen_dir, frozen_file ,optimised_dir,optimised_file, transforms, "dnn/head/predictions/class_ids")
describe_graph(get_graph_def_from_file(optimised_dir+optimised_file))
get_size(optimised_dir,optimised_file)
print("*"*100)
print("Convert back to saved model...")
export_dir="final_model/1/"
graph_file="optimised/1/saved_model.pb"
convert_graph_def_to_saved_model(export_dir, graph_file)
它“正常运行”，我的模型现在看起来像：
MetaGraphDef with tag-set: 'serve' contains the following SignatureDefs:
signature_def['serving_default']:
The given SavedModel SignatureDef contains the following input(s):
inputs['headline_placeholder'] tensor_info:
dtype: DT_STRING
shape: (-1)
name: headline_placeholder:0
The given SavedModel SignatureDef contains the following output(s):
outputs['class_ids'] tensor_info:
dtype: DT_INT64
shape: (-1)
name: dnn/head/predictions/class_ids:0
Method name is: tensorflow/serving/predict
我之所以说“有效”是因为我必须使用TF 1.14.1-dev20190516。否则存在数据类型问题，资产文件等事实。此外，如果没有remove_attribute(attribute_name=_class)行，则存在“预期与未知节点错误共存”的问题。
好的，现在我有128 MB的saved_model.pb，我尝试将其上传到AI Platform。
但是，当我这样做时，会收到错误消息
Create Version failed. Bad model detected with error: Model requires more memory than allowed. Please try to decrease the model size and re-deploy. If you continue to have error, please contact Cloud ML.
因此，在这一点上，我对如何上传此模型一无所知。有什么想法吗？