python sdk for metaai.py pipeline components
Project description
Python Metaai Pipeline Components SDK
快速开始
安装python-metaai
pip install -U python-metaai
使用说明
kubeflow pipeline component
编写组件代码demo
from typing import *
from metaai.pipeline.components import component,upload_model,image_datasets
from metaai.pipeline import pipeline
from metaai.pipeline.components.constants import ImagePullPolicy
from kfp.components import InputPath, OutputPath
@component(
# 如果您需要安装部分包,可以在这里进行指定,当然我们建议只安装体量较小的包,
# 如果您有自己独有的个性化包,请按照镜像编译手册来构建您的镜像
packages_to_install=["dill==0.3.5.1","requests==2.28.1"],
# enable_cache 指定当前的op是否会使用缓存。
# 如果没有改变我们建议打开缓存开关,默认为true
enable_cache=False,
)
def training(datasets_path: InputPath(str), training_config: Dict, model_path: OutputPath(str)):
import json
import requests
# dill 是pickle的再封装升级版本。可以dump 闭包属性
import dill
# 因为我需要dump当前这整个类 所以我引入了dill
class MM:
def fit(self,x):
setattr(self,"X",x)
def predict(self):
return {"prediction":self.X }
with open(datasets_path,"r") as fp:
datasets = json.load(fp)
items = datasets.get("items")
if not items and not isinstance(items,list):
raise ValueError("datasets input error! now datasets is {datasets}")
for item in items:
image_data = requests.get(item["image_url"]).content
print(len(image_data))
label_data = requests.get(item["label_url"]).content
print(len(label_data))
with open(model_path,"wb") as model_fp:
dill.dump({"model":MM(),"training_config":training_config},model_fp)
@pipeline(
name="pipeline-test",
description="python-metaai test pipeline",
# 如果你需要让每次这个容器运行都去重新拉取新的镜像,请将他为always
# 如:当您的基础镜像总是发生改变时,但是pipeline的代码却没有变动的时候。可以指定为always
image_pull_policy=ImagePullPolicy.Always.value,
)
# 你的pipeline将会接受的参数。
# 通过指定默认值,可以在ui界面上渲染出默认的key-value
def main(
dataset: Dict={"dataFiles":[],"labelFiles":[]},
train: Dict={},
model: Dict={"modelName":"model-test"},
):
datasets_op = image_datasets(dataset, enable_cache=False)
train_op = training(datasets=datasets_op.outputs["datasets"], training_config=train)
# train_op.set_gpu_limit(1)
# 被outputPath 或 InputPath等修饰的参数,会将后缀'_file'或'_path'进行裁剪,所以这类里直接使用outputs["model"]
# 为避免歧义,python-metaai 中都不直接使用return来在op间传递数据。
upload_model(train_op.outputs["model"], model, {"lalala": "lalala"})
if __name__ == "__main__":
my_first_pipeline = main()
# to yaml
my_first_pipeline.to_yaml("zt-sdk-test")
编写预测镜像
目录结构需要遵循以下目录树
. project_dir
├── my_module # 您的模块目录
│ ├── __init__.py
│ ├── main.py
│ └── pkgxxx
├── requirements.txt # 这个文件需要和serving.py同级,并且必须要存在,且可以为空
└── serving.py # 这个文件必须要存在,且和您的模块目录同级
serving代码demo,就demo pipeline生成的model
import dill
from typing import *
# 从metaai包中导入这些
from metaai.serving.models.commons.model import Model
from metaai.serving.models.commons.model_server import ModelServer
from metaai.serving.utils import success_response, failed_response_with_break
# 这里的导包需要注意,请不要使用相对路径导包
# 请直接从您的module顶层模块进行导入。
from my_module.main import preprocess_func
class MyModelService(Model):
# 如需重写 init 请super调用父类的init构造方法 保证基础实例子变量存在
def __init__(self, name: str):
super().__init__(name)
def _validate(self, request) -> Any:
# 在preprocess之后调用
# 这函数可以输入参数做自定义的校验。
# 比如您需要对这个request有相应的要求
if "balabala" not in request:
failed_response_with_break(message="expected key 'balabala' in request json!")
return super()._validate(request)
# 您可以使用async修饰当前函数, 也可以使用普通函数方法
async def _preprocess(self, request: Any) -> Any:
# 数据前处理的函数。
# 使用后会调用_validate
resp = await preprocess_func(request)
return resp
def load(self) -> bool:
# 使用适合模型的load方式
print("loading.....")
print(self.model_path)
with open(self.model_path,"rb") as fp:
pkl_res = dill.load(fp)
print(pkl_res)
self._model = pkl_res["model"]
self.ready = True
return self.ready
def predict(self, request: Any):
print(request)
# 数据预测函数
print("在predict")
self._model.fit(request)
return self._model.predict()
def _postprocess(self, response: Any) -> Dict:
# 数据后处理的函数。在predict后调用
return success_response(super()._postprocess(response))
# 启动服务需要 请务必写上
if __name__ == '__main__':
model = MyModelService(name="custom")
ModelServer.start(model)
使用命令编译docker 镜像并推送
# activate "虚拟环境"
metaai --help
cd "你的工作目录"
metaai serving-build -t {image_name}:{image_tag} -s .
# docker images 可以查看到刚刚编译出来的镜像
# docker login 进行登录
# docker push {image_name}:{image_tag} 推送镜像到远程
本地调试和使用
-
本地代码运行
# cd "你的工作目录" export MODEL_PATH=/home/{user1}/models/model.pkl python3 serving.py
容器运行
# 运行使用命令编译出来的 docker metaai镜像 # 比如说 image full name 是 metaai-serving:0.0.1 docker run -it --name metaai-serving -v /home/{user1}/models/model.pkl:/mnt/models/model.pkl \ -p 8089:8089 -e MODEL_PATH=/mnt/models/model.pkl metaai-serving:0.0.1
-
Output
# OUTPUT loading..... /home/zt/Workspace/demo-serving/model.pkl {'model': <class '__main__.tt.<locals>.MM'>} INFO: Started server process [120675] INFO: Waiting for application startup. INFO: Application startup complete. INFO: Uvicorn running on http://0.0.0.0:8089 (Press CTRL+C to quit)
调用方式
- 请求predict/ 接口
curl --request POST \
--url http://127.0.0.1:8089/predict/ \
--header 'Content-Type: application/json' \
--data '{
"body":"123"
}' -vv
- Output
# ouput
Note: Unnecessary use of -X or --request, POST is already inferred.
* Trying 127.0.0.1:8089...
* TCP_NODELAY set
* Connected to 127.0.0.1 (127.0.0.1) port 8089 (#0)
> POST /predict/ HTTP/1.1
> Host: 127.0.0.1:8089
> User-Agent: curl/7.68.0
> Accept: */*
> Content-Type: application/json
> Content-Length: 17
>
* upload completely sent off: 17 out of 17 bytes
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< date: Mon, 01 Aug 2022 10:39:04 GMT
< server: uvicorn
< content-length: 29
< content-type: application/json
<
* Connection #0 to host 127.0.0.1 left intact
{"prediction":{"body":"123"}}
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
python-metaai-0.1.0.dev5.tar.gz
(19.1 kB
view details)
Built Distribution
File details
Details for the file python-metaai-0.1.0.dev5.tar.gz
.
File metadata
- Download URL: python-metaai-0.1.0.dev5.tar.gz
- Upload date:
- Size: 19.1 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.1 CPython/3.8.5
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 26644fa090b31970fb3af7180f54f79e469f4bfe038b75f31190a1583dcd1ee3 |
|
MD5 | f7d43b1ca3d9093ac7f9469338eac4ad |
|
BLAKE2b-256 | ed3f6faadeefc64158fbb080e455c494e0ee06f89f887661696f00e79cc347fc |
File details
Details for the file python_metaai-0.1.0.dev5-py3-none-any.whl
.
File metadata
- Download URL: python_metaai-0.1.0.dev5-py3-none-any.whl
- Upload date:
- Size: 24.3 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.1 CPython/3.8.5
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 4c752f06ee85b6279478aa74ee81841c52a84b653c1cede8c9d2da4044996f9a |
|
MD5 | d30b7b17153bf15d31de59a97b82b29f |
|
BLAKE2b-256 | b0c50a8ee464da73a6b93a8fc86a93bf2b29f54e2d6de6de4571d45c4f6e1830 |