在本地机器调试 自定义引擎的规范可以在安装有docker的本地机器上通过以下步骤提前验证： 将自定义引擎镜像下载至本地机器，假设镜像名为custom_engine:v1。 将模型包文件夹复制到本地机器，假设模型包文件夹名字为model。 在模型包文件夹的同级目录下验证如下命令拉起服务： docker run --user 1000:100 -p 8080:8080 -v model:/home/mind/model custom_engine:v1 该指令无法完全模拟线上，主要是由于-v挂载进去的目录是root权限。在线上，模型文件从OBS下载到/home/mind/model目录之后，文件owner将统一修改为ma-user。 在本地机器上启动另一个终端，执行以下验证指令，得到符合预期的推理结果。 curl https://127.0.0.1:8080/${推理服务的请求路径}

https示例 使用Flask启动https，Webserver代码示例如下： from flask import Flask, request import json app = Flask(__name__) @app.route('/greet', methods=['POST']) def say_hello_func(): print("----------- in hello func ----------") data = json.loads(request.get_data(as_text=True)) print(data) username = data['name'] rsp_msg = 'Hello, {}!'.format(username) return json.dumps({"response":rsp_msg}, indent=4) @app.route('/goodbye', methods=['GET']) def say_goodbye_func(): print("----------- in goodbye func ----------") return '\nGoodbye!\n' @app.route('/', methods=['POST']) def default_func(): print("----------- in default func ----------") data = json.loads(request.get_data(as_text=True)) return '\n called default func !\n {} \n'.format(str(data)) @app.route('/health', methods=['GET']) def healthy(): return "{\"status\": \"OK\"}" # host must be "0.0.0.0", port must be 8080 if __name__ == '__main__': app.run(host="0.0.0.0", port=8080, ssl_context='adhoc')

MindSpore的推理脚本示例 snt3芯片目前只有北京四提工单申请权限后才可以使用，支持模型格式为.om，推理脚本如下： from __future__ import absolute_import from __future__ import division from __future__ import print_function import json import os import numpy as np from PIL import Image from hiai.nn_tensor_lib import NNTensor from hiai.nntensor_list import NNTensorList from model_service.hiai_model_service import HiaiBaseService class DemoService(HiaiBaseService): def __init__(self, *args, **kwargs): # 默认加载模型包目录下的om文件 super(DemoService, self).__init__(*args, **kwargs) self.labels_list = None self.is_multilabel = False def _preprocess(self, data): preprocessed_data = {} images = [] for k, v in data.items(): for file_name, file_content in v.items(): image = Image.open(file_content) image = np.array(image) # NHWC # AIPP should use RGB format. # mean reg is applied in AIPP. # Transpose is applied in AIPP tensor = NNTensor(image) images.append(tensor) tensor_list = NNTensorList(images) preprocessed_data['images'] = tensor_list return preprocessed_data def _inference(self, data, image_info=None): result = {} for k, v in data.items(): result[k] = self.model.proc(v) return result def _postprocess(self, data): # 这里增加自己的后处理 return str(data)

自定义推理逻辑的推理脚本示例 首先，需要在配置文件中，定义自己的依赖包，详细示例请参见使用自定义依赖包的模型配置文件示例。然后通过如下示例代码，实现了“saved_model”格式模型的加载推理。 当前推理基础镜像使用的python的logging模块，采用的是默认的日志级别Warning，即当前只有warning级别的日志可以默认查询出来。如果想要指定INFO等级的日志能够查询出来，需要在代码中指定logging的输出日志等级为INFO级别。

TensorFlow的推理脚本示例 TensorFlow MnistService示例如下。更多TensorFlow推理代码示例请参考Tensorflow、Tensorflow2.1。 推理代码 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 from PIL import Image import numpy as np from model_service.tfserving_model_service import TfServingBaseService class MnistService(TfServingBaseService): def _preprocess(self, data): preprocessed_data = {} for k, v in data.items(): for file_name, file_content in v.items(): image1 = Image.open(file_content) image1 = np.array(image1, dtype=np.float32) image1.resize((1, 784)) preprocessed_data[k] = image1 return preprocessed_data def _postprocess(self, data): infer_output = {} for output_name, result in data.items(): infer_output["mnist_result"] = result[0].index(max(result[0])) return infer_output 请求 curl -X POST \ 在线服务地址 \ -F images=@test.jpg 返回 {"mnist_result": 7} 在上面的代码示例中，完成了将用户表单输入的图片的大小调整，转换为可以适配模型输入的shape。首先通过Pillow库读取“32×32”的图片，调整图片大小为“1×784”以匹配模型输入。在后续处理中，转换模型输出为列表，用于Restful接口输出展示。

机器学习类型的模型配置文件示例 以下代码以XGBoost为例。 模型输入： { "req_data": [ { "sepal_length": 5, "sepal_width": 3.3, "petal_length": 1.4, "petal_width": 0.2 }, { "sepal_length": 5, "sepal_width": 2, "petal_length": 3.5, "petal_width": 1 }, { "sepal_length": 6, "sepal_width": 2.2, "petal_length": 5, "petal_width": 1.5 } ] } 模型输出： { "resp_data": [ { "predict_result": "Iris-setosa" }, { "predict_result": "Iris-versicolor" } ] } 配置文件： { "model_type": "XGBoost", "model_algorithm": "xgboost_iris_test", "runtime": "python2.7", "metrics": { "f1": 0.345294, "accuracy": 0.462963, "precision": 0.338977, "recall": 0.351852 }, "apis": [ { "url": "/", "method": "post", "request": { "Content-type": "application/json", "data": { "type": "object", "properties": { "req_data": { "items": [ { "type": "object", "properties": {} } ], "type": "array" } } } }, "response": { "Content-type": "applicaton/json", "data": { "type": "object", "properties": { "resp_data": { "type": "array", "items": [ { "type": "object", "properties": { "predict_result": {} } } ] } } } } } ] }

使用自定义依赖包的模型配置文件示例 如下示例中，定义了1.16.4版本的numpy的依赖环境。 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 { "model_algorithm": "image_classification", "model_type": "TensorFlow", "runtime": "python3.6", "apis": [ { "url": "/", "method": "post", "request": { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } }, "response": { "Content-type": "applicaton/json", "data": { "type": "object", "properties": { "mnist_result": { "type": "array", "item": [ { "type": "string" } ] } } } } } ], "metrics": { "f1": 0.124555, "recall": 0.171875, "precision": 0.00234938928519385, "accuracy": 0.00746268656716417 }, "dependencies": [ { "installer": "pip", "packages": [ { "restraint": "EXACT", "package_version": "1.16.4", "package_name": "numpy" } ] } ] }

图像分类模型配置文件示例 如下代码以TensorFlow引擎为例，您可以根据实际使用的引擎类型修改model_type参数后使用。 模型输入 key：images value：图片文件 模型输出 1 2 3 4 5 6 7 { "predicted_label": "flower", "scores": [ ["rose", 0.99], ["begonia", 0.01] ] } 配置文件 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 { "model_type": "TensorFlow", "model_algorithm": "image_classification", "metrics": { "f1": 0.345294, "accuracy": 0.462963, "precision": 0.338977, "recall": 0.351852 }, "apis": [{ "url": "/", "method": "post", "request": { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } }, "response": { "Content-type": "application/json", "data": { "type": "object", "properties": { "predicted_label": { "type": "string" }, "scores": { "type": "array", "items": [{ "type": "array", "minItems": 2, "maxItems": 2, "items": [ { "type": "string" }, { "type": "number" } ] }] } } } } }], "dependencies": [{ "installer": "pip", "packages": [{ "restraint": "ATLEAST", "package_version": "1.15.0", "package_name": "numpy" }, { "restraint": "", "package_version": "", "package_name": "Pillow" } ] }] } 如下代码以MindSpore引擎为例，您可以根据实际使用的引擎类型修改model_type参数后使用。 模型输入 key：images value：图片文件 模型输出 1 "[[-2.404526 -3.0476532 -1.9888215 0.45013925 -1.7018927 0.40332815\n -7.1861157 11.290332 -1.5861531 5.7887416 ]]" 配置文件 { "model_algorithm": "image_classification", "model_type": "MindSpore", "metrics": { "f1": 0.124555, "recall": 0.171875, "precision": 0.0023493892851938493, "accuracy": 0.00746268656716417 }, "apis": [{ "url": "/", "method": "post", "request": { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } }, "response": { "Content-type": "applicaton/json", "data": { "type": "object", "properties": { "mnist_result": { "type": "array", "item": [{ "type": "string" }] } } } } } ], "dependencies": [] }

预测分析模型配置文件示例 如下代码以TensorFlow引擎为例，您可以根据实际使用的引擎类型修改model_type参数后使用。 模型输入 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 { "data": { "req_data": [ { "buying_price": "high", "maint_price": "high", "doors": "2", "persons": "2", "lug_boot": "small", "safety": "low", "acceptability": "acc" }, { "buying_price": "high", "maint_price": "high", "doors": "2", "persons": "2", "lug_boot": "small", "safety": "low", "acceptability": "acc" } ] } } 模型输出 1 2 3 4 5 6 7 8 9 10 11 12 { "data": { "resp_data": [ { "predict_result": "unacc" }, { "predict_result": "unacc" } ] } } 配置文件 代码中request结构和response结构中的data参数是json schema数据结构。data/properties里面的内容对应“模型输入”和“模型输出”。 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 { "model_type": "TensorFlow", "model_algorithm": "predict_analysis", "metrics": { "f1": 0.345294, "accuracy": 0.462963, "precision": 0.338977, "recall": 0.351852 }, "apis": [ { "url": "/", "method": "post", "request": { "Content-type": "application/json", "data": { "type": "object", "properties": { "data": { "type": "object", "properties": { "req_data": { "items": [ { "type": "object", "properties": {} } ], "type": "array" } } } } } }, "response": { "Content-type": "application/json", "data": { "type": "object", "properties": { "data": { "type": "object", "properties": { "resp_data": { "type": "array", "items": [ { "type": "object", "properties": {} } ] } } } } } } } ], "dependencies": [ { "installer": "pip", "packages": [ { "restraint": "EXACT", "package_version": "1.15.0", "package_name": "numpy" }, { "restraint": "EXACT", "package_version": "5.2.0", "package_name": "Pillow" } ] } ] }

自定义镜像 类型的模型配置文件示例 模型输入和输出与目标检测模型配置文件示例类似。 模型预测输入为图片类型时，request请求示例如下： 该示例表示模型预测接收一个参数名为images、参数类型为file的预测请求，在推理界面会显示文件上传按钮，以文件形式进行预测。 1 2 3 4 5 6 7 8 9 10 11 { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } } 模型预测输入为json数据类型时，request请求示例如下： 该示例表示模型预测接收json请求体，只有一个参数名为input、参数类型为string的预测请求，在推理界面会显示文本输入框，用于填写预测请求。 1 2 3 4 5 6 7 8 9 10 11 { "Content-type": "application/json", "data": { "type": "object", "properties": { "input": { "type": "string" } } } } 完整请求示例如下： 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 { "model_algorithm": "image_classification", "model_type": "Image", "metrics": { "f1": 0.345294, "accuracy": 0.462963, "precision": 0.338977, "recall": 0.351852 }, "apis": [{ "url": "/", "method": "post", "request": { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } }, "response": { "Content-type": "application/json", "data": { "type": "object", "required": [ "predicted_label", "scores" ], "properties": { "predicted_label": { "type": "string" }, "scores": { "type": "array", "items": [{ "type": "array", "minItems": 2, "maxItems": 2, "items": [{ "type": "string" }, { "type": "number" } ] }] } } } } }] }

apis参数代码示例 [{ "url": "/", "method": "post", "request": { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } }, "response": { "Content-type": "applicaton/json", "data": { "type": "object", "properties": { "mnist_result": { "type": "array", "item": [ { "type": "string" } ] } } } } }]

目标检测模型配置文件示例 如下代码以TensorFlow引擎为例，您可以根据实际使用的引擎类型修改model_type参数后使用。 模型输入 key：images value：图片文件 模型输出 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 { "detection_classes": [ "face", "arm" ], "detection_boxes": [ [ 33.6, 42.6, 104.5, 203.4 ], [ 103.1, 92.8, 765.6, 945.7 ] ], "detection_scores": [0.99, 0.73] } 配置文件 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 { "model_type": "TensorFlow", "model_algorithm": "object_detection", "metrics": { "f1": 0.345294, "accuracy": 0.462963, "precision": 0.338977, "recall": 0.351852 }, "apis": [{ "url": "/", "method": "post", "request": { "Content-type": "multipart/form-data", "data": { "type": "object", "properties": { "images": { "type": "file" } } } }, "response": { "Content-type": "application/json", "data": { "type": "object", "properties": { "detection_classes": { "type": "array", "items": [{ "type": "string" }] }, "detection_boxes": { "type": "array", "items": [{ "type": "array", "minItems": 4, "maxItems": 4, "items": [{ "type": "number" }] }] }, "detection_scores": { "type": "array", "items": [{ "type": "number" }] } } } } }], "dependencies": [{ "installer": "pip", "packages": [{ "restraint": "EXACT", "package_version": "1.15.0", "package_name": "numpy" }, { "restraint": "EXACT", "package_version": "5.2.0", "package_name": "Pillow" } ] }] }