通过注解构图并执行

import dataflow as df
import numpy as np

@df.pyflow
def add(in0, in1):
    return in0 + in1

# 系统初始化
options = {
    "ge.exec.deviceId":"0",
    "ge.experiment.data_flow_deploy_info_path":"./data_flow_deploy_info.json",
    "ge.socVersion": "AscendXXX"  # 根据环境修改version
}
df.init(options)

# 定义FlowData
data0 = df.FlowData()
data1 = df.FlowData()
# 通过注解UDF生成FlowNode
add_node = add.fnode()

# 构建FlowData和FlowNode的连边关系
add_out = add_node(data0, data1)

# 通过FlowOut构建FlowGraph
dag = df.FlowGraph([add_out])

# 根据图自动生成部署策略，所有节点均部署在numa_config中的第1个设备上
# 如需修改需要注释下面这行代码，并修改./data_flow_deploy_info.json文件内容
df.utils.generate_deploy_template(dag, "./data_flow_deploy_info.json")

# 调用FlowGraph.feed填充输入
dag.feed({data0:np.array([[1, 2]], dtype=np.int32), data1:np.array([[2, 3]], dtype=np.int32)})

# 调用FlowGraph.fetch获取输出
print("dataflow fetch result:", dag.fetch())
# 打印：dataflow fetch result: ([array([[3, 5]], dtype=int32)], 0)

# 释放系统资源 
df.finalize()