BERT

This page illustrates a simple client-server interaction to perform inference on a TensorFlow model using the Java SDK for Konduit Serving.

import ai.konduit.serving.InferenceConfiguration;
import ai.konduit.serving.config.ParallelInferenceConfig;
import ai.konduit.serving.config.ServingConfig;
import ai.konduit.serving.configprovider.KonduitServingMain;
import ai.konduit.serving.configprovider.KonduitServingMainArgs;
import ai.konduit.serving.model.ModelConfig;
import ai.konduit.serving.model.ModelConfigType;
import ai.konduit.serving.model.TensorDataTypesConfig;
import ai.konduit.serving.model.TensorFlowConfig;
import ai.konduit.serving.pipeline.step.ModelStep;
import ai.konduit.serving.verticles.inference.InferenceVerticle;
import com.mashape.unirest.http.Unirest;
import com.mashape.unirest.http.exceptions.UnirestException;
import org.apache.commons.io.FileUtils;
import org.nd4j.linalg.io.ClassPathResource;
import org.nd4j.tensorflow.conversion.TensorDataType;

Overview

Konduit Serving works by defining a series of steps. These include operations such as

1. Pre- or post-processing steps

2. One or more machine learning models

3. Transforming the output in a way that can be understood by humans

If deploying your model does not require pre- nor post-processing, only one step - a machine learning model - is required. This configuration is defined using a single ModelStep.

Start by downloading the model weights to the data folder.The downloaded zip file can be unzipped using Util class(Util.unzipBertFile).

String bertmodelfilePath = new ClassPathResource("data/bert").getFile().getAbsolutePath();
String bertFileName = "bert_mrpc_frozen.pb";
File bertModelFile = new File(bertDataFolder, bertFileName);
File bertFile = new File(bertmodelfilePath);
if (!bertModelFile.exists()) {
    File bertDownloadedZipFile = Util.downloadBertModel();
    Util.unzipBertFile(bertDownloadedZipFile.toString(), bertFileName);
}

A reference Java project is provided in the Example repository ( https://github.com/KonduitAI/konduit-serving-examples ) with a Maven pom.xml dependencies file. If using the IntelliJ IDEA IDE, open the java folder as a Maven project and run the main function of InferenceModelStepBERT the class.

Configure the step

Define the TensorFlow configuration as a TensorFlowConfig object.

• tensorDataTypesConfig: The TensorFlowConfig object requires a HashMap input_data_types. Its keys should represent column names, and the values should represent data types as strings, e.g. "INT32". See here for a list of supported data types.

• modelConfigType: This argument requires a ModelConfigType object. Specify modelType as TENSORFLOW, and modelLoadingPath to point to the location of TensorFlow weights saved in the PB file format.

HashMap<String, TensorDataType> input_data_types = new LinkedHashMap<>();
input_data_types.put("IteratorGetNext:0", TensorDataType.INT32);
input_data_types.put("IteratorGetNext:1", TensorDataType.INT32);
input_data_types.put("IteratorGetNext:4", TensorDataType.INT32);

ModelConfig bertModelConfig = TensorFlowConfig.builder()
    .tensorDataTypesConfig(TensorDataTypesConfig.builder().
            inputDataTypes(input_data_types).build())
    .modelConfigType(ModelConfigType.builder().
            modelLoadingPath(bertModelFile.getAbsolutePath()).
            modelType(ModelConfig.ModelType.TENSORFLOW).build())
    .build();

Now that we have a TensorFlowConfig defined, we can define a ModelStep. The following parameters are specified:

• modelConfig: pass the TensorFlowConfig object here

• parallelInferenceConfig: specify the number of workers to run in parallel. Here, we specify workers=1.

• inputNames: names for the input data

• outputNames: names for the output data

List<String> input_names = new ArrayList<String>(input_data_types.keySet());
ArrayList<String> output_names = new ArrayList<>();
output_names.add("loss/Softmax");

ModelStep bertModelStep = ModelStep.builder()
    .modelConfig(bertModelConfig)
    .inputNames(input_names)
    .outputNames(output_names)
    .parallelInferenceConfig(ParallelInferenceConfig.builder().workers(1).build())
    .build();

Configure the server

Specify the following:

• httpPort: specify any port number that is not reserved.

int port = Util.randInt(1000, 65535);
ServingConfig servingConfig = ServingConfig.builder().httpPort(port)
    .build();

The ServingConfig has to be passed to Server in addition to the steps as a Python list. In this case, there is a single step: bertModelStep.

InferenceConfiguration inferenceConfiguration = InferenceConfiguration.builder()
    .servingConfig(servingConfig)
    .step(bertModelStep)
    .build();

The inferenceConfiguration is stored as a JSON File. Set the KonduitServingMainArgs with the saved config.json file path as configPath and other necessary server configuration arguments.

File configFile = new File("config.json");
FileUtils.write(configFile, inferenceConfiguration.toJson(), Charset.defaultCharset());

KonduitServingMainArgs args1 = KonduitServingMainArgs.builder()
    .configStoreType("file").ha(false)
    .multiThreaded(false).configPort(port)
    .verticleClassName(InferenceVerticle.class.getName())
    .configPath(configFile.getAbsolutePath())
    .build();

Start server by calling KonduitServingMain with the configurations mentioned in the KonduitServingMainArgs using Callback Function(as per the code mentioned in the Inference Section below)

Inference

Load some sample data from NumPy files. Note that these are NumPy arrays, each with shape (4, 128):

File input0 = new ClassPathResource("data/bert/input-0.npy").getFile();
File input1 = new ClassPathResource("data/bert/input-1.npy").getFile();
File input4 = new ClassPathResource("data/bert/input-4.npy").getFile();

To configure the client, set the required URL to connect server and specify any port number that is not reserved (as used in server configuration).

A Callback Function onSuccess is implemented in order to post the Client request and get the HttpResponse, only after the successful run of the KonduitServingMain Server.

Accepted input and output data formats are as follows:

• Input: JSON, ARROW, IMAGE, ND4J (not yet implemented) and NUMPY.

• Output: NUMPY, JSON, ND4J (not yet implemented) and ARROW.

 KonduitServingMain.builder()
      .onSuccess(()->{
          try {
              String response = Unirest.post(String.format("http://localhost:%s/raw/numpy", port))
                      .field("IteratorGetNext:0", input0)
                      .field("IteratorGetNext:1", input1)
                      .field("IteratorGetNext:4", input4)
                      .asString().getBody();
              System.out.print(response);
              System.exit(0);
          } catch (UnirestException e) {
              e.printStackTrace();
              System.exit(0);
          }
      })
      .build()
      .runMain(args1.toArgs());

Confirm the output

After executing the above, in order to confirm the successful start of the Server, check for the below output text:

Jan 07, 2020 6:02:49 PM ai.konduit.serving.configprovider.KonduitServingMain
INFO: Deployed verticle ai.konduit.serving.verticles.inference.InferenceVerticle

The Output of the program is as follows:

System.out.print(response);

"loss/Softmax" : {
  "batchId" : "41600218-5fb7-401f-af7d-e7fe13313f5d",
  "ndArray" : {
    "dataType" : "FLOAT",
    "shape" : [ 4, 2 ],
    "data" : [ 0.9894917, 0.010508226, 0.8021635, 0.19783656, 0.9874369, 0.012563077, 0.99294597, 0.0070540793 ]
  }

The complete inference configuration in JSON format is as follows:

System.out.println(inferenceConfiguration.toJson());

{
  "memMapConfig" : null,
  "servingConfig" : {
    "httpPort" : 11805,
    "inputDataFormat" : "NUMPY",
    "listenHost" : "localhost",
    "logTimings" : false,
    "metricTypes" : [ "CLASS_LOADER", "JVM_MEMORY", "JVM_GC", "PROCESSOR", "JVM_THREAD", "LOGGING_METRICS", "NATIVE" ],
    "outputDataFormat" : "JSON",
    "predictionType" : "RAW",
    "uploadsDirectory" : "file-uploads/"
  },
  "steps" : [ {
    "@type" : "ModelStep",
    "inputColumnNames" : { },
    "inputNames" : [ "IteratorGetNext:0", "IteratorGetNext:1", "IteratorGetNext:4" ],
    "inputSchemas" : { },
    "modelConfig" : {
      "@type" : "TensorFlowConfig",
      "configProtoPath" : null,
      "modelConfigType" : {
        "modelLoadingPath" : "C:\\konduit-serving-examples\\java\\target\\classes\\data\\bert\\bert_mrpc_frozen.pb",
        "modelType" : "TENSORFLOW"
      },
      "savedModelConfig" : null,
      "tensorDataTypesConfig" : {
        "inputDataTypes" : {
          "IteratorGetNext:0" : "INT32",
          "IteratorGetNext:1" : "INT32",
          "IteratorGetNext:4" : "INT32"
        },
        "outputDataTypes" : { }
      }
    },
    "normalizationConfig" : null,
    "outputColumnNames" : { },
    "outputNames" : [ "loss/Softmax" ],
    "outputSchemas" : { },
    "parallelInferenceConfig" : {
      "batchLimit" : 32,
      "inferenceMode" : "BATCHED",
      "maxTrainEpochs" : 1,
      "queueLimit" : 64,
      "vertxConfigJson" : null,
      "workers" : 1
    }
  } ]
}