matreche/modelarts-umn
1
0
Fork 0
forked from docs/modelarts
Code Issues Pull Requests Packages Projects Releases Wiki Activity
modelarts-umn/api-ref/source/model_management/importing_a_model.rst

47 KiB
Raw Blame History

original_name

modelarts_03_0076.html

Importing a Model

Function

You can use the API to import a model.

Ensure that the execution code and model have been uploaded to OBS. By default, the models generated by a training job are stored in OBS.

URI

POST /v1/{project_id}/models

Table 1 <modelarts_03_0076__en-us_topic_0130147365_table16518993181628> describes the required parameters.

Table 1 Parameters
Parameter Mandatory Type Description
project_id Yes String Project ID. For details about how to obtain the project ID, see Obtaining a Project ID <modelarts_03_0147>.

Request Body

Table 2 <modelarts_03_0076__en-us_topic_0130147365_table24514577587> describes the request parameters.

Table 2 Parameters
Parameter Mandatory Type Description
model_name Yes String Model name. Enter 1 to 64 characters. Only letters, digits, hyphens (-), and underscores (_) are allowed.
model_version Yes String Model version in the format of Digit.Digit.Digit. The value range of the digits is [1, 99]. Note that no part of the version number can start with 0. For example, 01.01.01 is not allowed.
source_location Yes String

OBS path where the model is located or the SWR image location.

  • If model_type is set to TensorFlow, MXNet, Caffe, Spark_MLlib, Scikit_Learn, XGBoost, PyTorch, or Template, set source_location to the OBS path where the model is located.
  • If model_type is set to Image, set source_location to the path where the SWR image is located.
source_job_id No String ID of the source training job. If the model is generated from a training job, input this parameter for source tracing. If the model is imported from a third-party meta model, leave this parameter blank. By default, this parameter is left blank.
source_job_version No String Version of the source training job. If the model is generated from a training job, input this parameter for source tracing. If the model is imported from a third-party meta model, leave this parameter blank. By default, this parameter is left blank.
source_type No String Model source type. The value can only be auto, which indicates ExeML models (model download is not supported). If a model is deployed through a training job, this parameter is left blank by default.
model_type Yes String Model type. The value can be TensorFlow, MXNet, Caffe, Spark_MLlib, Scikit_Learn, XGBoost, Image, PyTorch, or Template, which is read from the configuration file.
runtime No String For details about runtime options, see Model Management > Importing a Model > Importing a Meta Model from OBS in ModelArts User Guide.
description No String Model remarks. The value contains a maximum of 100 characters and cannot contain the following special characters and more: &!'\"<>=
model_algorithm No String Model algorithm. If the algorithm is read from the configuration file, this parameter can be left blank. For example, the value can be predict_analysis, object_detection, or image_classification. The value must start with a letter and contain no more than 36 characters.
execution_code No String OBS path for storing the execution code. By default, this parameter is left blank. The name of the execution code file is fixed to customize_service.py. The inference code file must be stored in the model directory. If this parameter is left blank, the system can automatically identify the inference code in the model directory.
input_params No params array Collection of input parameters of a model. By default, this parameter is left blank. For details, see Table 3 <modelarts_03_0076__en-us_topic_0130147365_table01082501075>. If the collection of input parameters is read from apis in the configuration file, you only need to provide the initial_config field and do not need to set input_params. input_params is optional when the initial_config field exists.
output_params No params array Collection of output parameters of a model. By default, this parameter is left blank. For details, see Table 3 <modelarts_03_0076__en-us_topic_0130147365_table01082501075>. If the collection of output parameters is read from apis in the configuration file, you only need to provide the initial_config field and do not need to set output_params.
dependencies No dependency array Package required for inference code and model. By default, this parameter is left blank. If the package is read from the configuration file, this parameter can be left blank. Table 4 <modelarts_03_0076__en-us_topic_0130147365_table649013511785> shows the dependency structure.
model_metrics No String Model precision, which is read from the configuration file
apis No apis array All apis input and output parameters of the model. If the parameters are read from the configuration file, this parameter can be left blank. For details, see Table 10 <modelarts_03_0076__en-us_topic_0130147365_table53974919117>.
initial_config No String Character string converted from the final model configuration file. It is recommended that the initial_config file be used to provide information about the fields such as apis, dependencies, input_params, and output_params.
workspace_id No String ID of the workspace to which a service belongs. The default value is 0, indicating the default workspace.
model_docs No doc array List of model description documents. A maximum of three documents are supported. For details, see Table 7 <modelarts_03_0076__en-us_topic_0130147365_table9739172671118>.
install_type No String array Deployment type. Only lowercase letters are supported. The value can be real-time, or batch. Default value: [real-time, batch]
template No Template object Template configuration items. This parameter is mandatory when model_type is set to Template. For details, see Table 8 <modelarts_03_0076__en-us_topic_0130147365_table89901124131016>.
Table 3 params parameters
Parameter Mandatory Type Description
url Yes String API URL
method Yes String Request method. Possible values are post and get.
protocol Yes String Request protocol
param_name Yes String Parameter name. It is recommended that the parameter name contain a maximum of 64 characters.
param_type Yes String Parameter type. The value can be int, string, float, timestamp, date, or file.
min No Number This parameter is optional when param_type is set to int or float. By default, this parameter is left blank.
max No Number This parameter is optional when param_type is set to int or float. By default, this parameter is left blank.
param_desc No String Parameter description. It is recommended that the parameter description contain a maximum of 100 characters. By default, this parameter is left blank.
Table 4 dependency parameters
Parameter Mandatory Type Description
installer Yes String Installation mode. Only pip is supported.
packages Yes package array Collection of dependency packages. For details, see Table 5 <modelarts_03_0076__en-us_topic_0130147365_table137621937181116>.
Table 5 package parameters
Parameter Mandatory Type Description
package_name Yes String Name of a dependency package. Ensure that the package name is correct and exists.
package_version No String Version of a dependency package. If this parameter is left blank, the latest version is installed by default.
restraint No String Version restriction. The value can be EXACT, ATLEAST, or ATMOST. This parameter is mandatory only when package_version exists.
Table 6 metric parameters
Parameter Mandatory Type Description
f1 No Double F1 score
recall No Double Recall
precision No Double Precision
accuracy No Double Accuracy
Table 7 doc parameters
Parameter Mandatory Type Description
doc_name Yes String Document name, which must start with a letter. Enter 1 to 48 characters. Only letters, digits, hyphens (-), and underscores (_) are allowed.
doc_url Yes String HTTP(S) link of the document
Table 8 Template parameters
Parameter Mandatory Type Description
infer_format No String Input and output mode. When this parameter is used, the input and output mode built in the template does not take effect.
template_id Yes String ID of the used template. The template has a built-in input and output mode.
template_inputs Yes Array of TemplateInputs objects Template input configuration, specifying the source path for configuring a model. For details, see Table 9 <modelarts_03_0076__en-us_topic_0130147365_table12364181131310>.
Table 9 template_inputs parameters
Parameter Mandatory Type Description
input Yes String Template input path, which can be an OBS file path or OBS directory path. When you use a template with multiple input items to create a model, if the target paths input_properties specified in the template are the same, the OBS directory or OBS file name entered here must be unique to prevent files from being overwritten.
input_id Yes String Input item ID, which is obtained from the template details.
Table 10 apis parameters
Parameter Mandatory Type Description
input_params No Object Input parameters in apis, described in JSON Schema format. For details, see Table 11 <modelarts_03_0076__en-us_topic_0130147365_table116145629>.
method No String Request method. The options are POST and GET.
output_params No Object Output parameters in apis, described in JSON Schema format. For details, see Table 12 <modelarts_03_0076__en-us_topic_0130147365_table890117461320>.
protocol No String Request protocol.
url No String Inference request URL.
Table 11 InputParams parameters
Parameter Mandatory Type Description
properties No Map<String,Object> Properties of an object element in JSON Schema. You can set parameters, including the parameter name and type, in properties.
type No String Type in JSON Schema, which can be object.
Table 12 OutputParams parameters
Parameter Mandatory Type Description
properties No Map<String,Object> Properties of an object element in JSON Schema. You can set parameters, including the parameter name and type, in properties.
type No String Type in JSON Schema, which can be object.

Response Body

Table 13 <modelarts_03_0076__en-us_topic_0130147365_table88391251102419> describes the response parameters.

Table 13 Parameters
Parameter Type Description
model_id String Model ID

Samples

The following shows how to import a model whose name is mnist, version is 1.0.0, and type is TensorFlow. The model file comes from an OBS bucket.

  • Sample request

    POST    https://endpoint/v1/{project_id}/models
{
"model_name": "mnist",
"model_version": "1.0.0",
"source_location": "https://models.obs.xxxx.com/mnist",
"source_job_id": "55",
"source_job_version": "V100",
"model_type": "TensorFlow",
"runtime": "python2.7",
"description": "mnist model",
"execution_code": "https://testmodel.obs.xxxx.com/customize_service.py",
"input_params": [
{
  "url": "/v1/xxx/image",
  "protocol": "http",
  "method": "post",
  "param_name": "image_url",
  "param_type": "string",
  "min": 0,
  "max": 9,
  "param_desc": "http://test/test.jpeg"
}
],
"output_params": [
{
  "url": "/v1/xxx/image",
  "protocol": "http",
  "method": "post",
  "param_name": "face_location",
  "param_type": "box",
  "param_desc": "face_location param value description"
}
],
"dependencies": [
{
  "installer": "pip",
  "packages": [
    {
      "package_name": "numpy",
      "package_version": "1.5.0",
      "restraint": "ATLEAST"
    }
  ]
}
],
"model_algorithm": "object_detection",
"model_metrics":"{\"f1\":0.52381,\"recall\":0.666667,\"precision\":0.466667,\"accuracy\":0.625}",
"apis": [
{
  "url": "/v1/xxx/image",
  "protocol": "http",
  "method": "post",
  "input_params": {
    "type": "object",
    "properties": {
      "image_url": {
        "type": "string"
      }
    }
  },
  "output_params": {
    "type": "object",
    "properties": {
      "face_location": {
        "type": "box"
      }
    }
  }
}
]
}

  • Sample response

    {
  "model_id": "10eb0091-887f-4839-9929-cbc884f1e20e"
}

Status Code

For details about the status code, see Table 1 <modelarts_03_0094__en-us_topic_0132773864_table1450010510213>.

Error Codes

See Error Codes <modelarts_03_0095>.