adding import notebook + changing default model + adding onnx support (…

…#14158)

python/sparknlp/annotator/classifier_dl/bert_for_zero_shot_classification.py

@@ -41,7 +41,7 @@ class BertForZeroShotClassification(AnnotatorModel,
     ...     .setInputCols(["token", "document"]) \\
     ...     .setOutputCol("label")
 
-    The default model is ``"bert_base_cased_zero_shot_classifier_xnli"``, if no name is
+    The default model is ``"bert_zero_shot_classifier_mnli"``, if no name is
     provided.
 
     For available pretrained models please see the `Models Hub
@@ -189,14 +189,14 @@ def loadSavedModel(folder, spark_session):
         return BertForZeroShotClassification(java_model=jModel)
 
     @staticmethod
-    def pretrained(name="bert_base_cased_zero_shot_classifier_xnli", lang="en", remote_loc=None):
+    def pretrained(name="bert_zero_shot_classifier_mnli", lang="xx", remote_loc=None):
         """Downloads and loads a pretrained model.
 
         Parameters
         ----------
         name : str, optional
             Name of the pretrained model, by default
-            "bert_base_cased_zero_shot_classifier_xnli"
+            "bert_zero_shot_classifier_mnli"
         lang : str, optional
             Language of the pretrained model, by default "en"
         remote_loc : str, optional

python/test/annotator/classifier_dl/bert_for_zero_shot_classification_test.py

@@ -29,7 +29,7 @@ def setUp(self):
             .toDF("text")
 
         self.tested_annotator = BertForZeroShotClassification \
-            .pretrained("bert_base_cased_zero_shot_classifier_xnli") \
+            .pretrained() \
             .setInputCols(["document", "token"]) \
             .setOutputCol("class") \
             .setCandidateLabels(["urgent", "mobile", "travel", "movie", "music", "sport", "weather", "technology"])

src/main/scala/com/johnsnowlabs/ml/ai/BertClassification.scala

@@ -257,6 +257,7 @@ private[johnsnowlabs] class BertClassification(
         embeddings
       } finally if (results != null) results.close()
     }
+
   }
 
   def tagSequence(batch: Seq[Array[Int]], activation: String): Array[Array[Float]] = {
@@ -284,14 +285,62 @@ private[johnsnowlabs] class BertClassification(
     batchScores
   }
 
-  def tagZeroShotSequence(
+  def computeZeroShotLogitsWithONNX(
       batch: Seq[Array[Int]],
-      entailmentId: Int,
-      contradictionId: Int,
-      activation: String): Array[Array[Float]] = {
-    val tensors = new TensorResources()
+      maxSentenceLength: Int): Array[Float] = {
 
-    val maxSentenceLength = batch.map(encodedSentence => encodedSentence.length).max
+    val (runner, env) = onnxWrapper.get.getSession(onnxSessionOptions)
+
+    val tokenTensors =
+      OnnxTensor.createTensor(env, batch.map(x => x.map(x => x.toLong)).toArray)
+    val maskTensors =
+      OnnxTensor.createTensor(
+        env,
+        batch.map(sentence => sentence.map(x => if (x == 0L) 0L else 1L)).toArray)
+
+    val segmentTensors =
+      OnnxTensor.createTensor(
+        env,
+        batch
+          .map(sentence =>
+            sentence.indices
+              .map(i =>
+                if (i < sentence.indexOf(sentenceEndTokenId)) 0L
+                else if (i == sentence.indexOf(sentenceEndTokenId)) 1L
+                else 1L)
+              .toArray)
+          .toArray)
+
+    val inputs =
+      Map(
+        "input_ids" -> tokenTensors,
+        "attention_mask" -> maskTensors,
+        "token_type_ids" -> segmentTensors).asJava
+
+    try {
+      val results = runner.run(inputs)
+      try {
+        val embeddings = results
+          .get("logits")
+          .get()
+          .asInstanceOf[OnnxTensor]
+          .getFloatBuffer
+          .array()
+        tokenTensors.close()
+        maskTensors.close()
+        segmentTensors.close()
+
+        embeddings
+      } finally if (results != null) results.close()
+    }
+
+  }
+
+  def computeZeroShotLogitsWithTF(
+      batch: Seq[Array[Int]],
+      maxSentenceLength: Int): Array[Float] = {
+
+    val tensors = new TensorResources()
     val batchLength = batch.length
 
     val tokenBuffers: IntDataBuffer = tensors.createIntBuffer(batchLength * maxSentenceLength)
@@ -350,6 +399,23 @@ private[johnsnowlabs] class BertClassification(
     tensors.clearSession(outs)
     tensors.clearTensors()
 
+    rawScores
+  }
+
+  def tagZeroShotSequence(
+      batch: Seq[Array[Int]],
+      entailmentId: Int,
+      contradictionId: Int,
+      activation: String): Array[Array[Float]] = {
+
+    val maxSentenceLength = batch.map(encodedSentence => encodedSentence.length).max
+    val batchLength = batch.length
+
+    val rawScores = detectedEngine match {
+      case ONNX.name => computeZeroShotLogitsWithONNX(batch, maxSentenceLength)
+      case _ => computeZeroShotLogitsWithTF(batch, maxSentenceLength)
+    }
+
     val dim = rawScores.length / batchLength
     rawScores
       .grouped(dim)

src/main/scala/com/johnsnowlabs/nlp/annotators/classifier/dl/BertForZeroShotClassification.scala

@@ -17,14 +17,14 @@
 package com.johnsnowlabs.nlp.annotators.classifier.dl
 
 import com.johnsnowlabs.ml.ai.BertClassification
-import com.johnsnowlabs.ml.onnx.OnnxWrapper
+import com.johnsnowlabs.ml.onnx.{OnnxWrapper, ReadOnnxModel, WriteOnnxModel}
 import com.johnsnowlabs.ml.tensorflow._
 import com.johnsnowlabs.ml.util.LoadExternalModel.{
   loadTextAsset,
   modelSanityCheck,
   notSupportedEngineError
 }
-import com.johnsnowlabs.ml.util.TensorFlow
+import com.johnsnowlabs.ml.util.{ONNX, TensorFlow}
 import com.johnsnowlabs.nlp._
 import com.johnsnowlabs.nlp.annotators.common._
 import com.johnsnowlabs.nlp.serialization.MapFeature
@@ -50,7 +50,7 @@ import org.apache.spark.sql.SparkSession
   *   .setInputCols("token", "document")
   *   .setOutputCol("label")
   * }}}
-  * The default model is `"bert_base_cased_zero_shot_classifier_xnli"`, if no name is provided.
+  * The default model is `"bert_zero_shot_classifier_mnli"`, if no name is provided.
   *
   * For available pretrained models please see the
   * [[https://sparknlp.org/models?task=Text+Classification Models Hub]].
@@ -124,6 +124,7 @@ class BertForZeroShotClassification(override val uid: String)
     extends AnnotatorModel[BertForZeroShotClassification]
     with HasBatchedAnnotate[BertForZeroShotClassification]
     with WriteTensorflowModel
+    with WriteOnnxModel
     with HasCaseSensitiveProperties
     with HasClassifierActivationProperties
     with HasEngine
@@ -338,21 +339,34 @@ class BertForZeroShotClassification(override val uid: String)
 
   override def onWrite(path: String, spark: SparkSession): Unit = {
     super.onWrite(path, spark)
-    writeTensorflowModelV2(
-      path,
-      spark,
-      getModelIfNotSet.tensorflowWrapper.get,
-      "_bert_classification",
-      BertForZeroShotClassification.tfFile,
-      configProtoBytes = getConfigProtoBytes)
+    val suffix = "_bert_classification"
+
+    getEngine match {
+      case TensorFlow.name =>
+        writeTensorflowModelV2(
+          path,
+          spark,
+          getModelIfNotSet.tensorflowWrapper.get,
+          suffix,
+          BertForZeroShotClassification.tfFile,
+          configProtoBytes = getConfigProtoBytes)
+      case ONNX.name =>
+        writeOnnxModel(
+          path,
+          spark,
+          getModelIfNotSet.onnxWrapper.get,
+          suffix,
+          BertForZeroShotClassification.onnxFile)
+    }
   }
 
 }
 
 trait ReadablePretrainedBertForZeroShotModel
     extends ParamsAndFeaturesReadable[BertForZeroShotClassification]
     with HasPretrained[BertForZeroShotClassification] {
-  override val defaultModelName: Some[String] = Some("bert_base_cased_zero_shot_classifier_xnli")
+  override val defaultModelName: Some[String] = Some("bert_zero_shot_classifier_mnli")
+  override val defaultLang: String = "xx"
 
   /** Java compliant-overrides */
   override def pretrained(): BertForZeroShotClassification = super.pretrained()
@@ -368,19 +382,29 @@ trait ReadablePretrainedBertForZeroShotModel
       remoteLoc: String): BertForZeroShotClassification = super.pretrained(name, lang, remoteLoc)
 }
 
-trait ReadBertForZeroShotDLModel extends ReadTensorflowModel {
+trait ReadBertForZeroShotDLModel extends ReadTensorflowModel with ReadOnnxModel {
   this: ParamsAndFeaturesReadable[BertForZeroShotClassification] =>
 
   override val tfFile: String = "bert_classification_tensorflow"
+  override val onnxFile: String = "bert_classification_onnx"
 
   def readModel(
       instance: BertForZeroShotClassification,
       path: String,
       spark: SparkSession): Unit = {
 
-    val tensorFlow =
-      readTensorflowModel(path, spark, "_bert_classification_tf", initAllTables = false)
-    instance.setModelIfNotSet(spark, Some(tensorFlow), None)
+    instance.getEngine match {
+      case TensorFlow.name =>
+        val tensorFlow =
+          readTensorflowModel(path, spark, "_bert_classification_tf", initAllTables = false)
+        instance.setModelIfNotSet(spark, Some(tensorFlow), None)
+      case ONNX.name =>
+        val onnxWrapper =
+          readOnnxModel(path, spark, "_bert_classification_onnx")
+        instance.setModelIfNotSet(spark, None, Some(onnxWrapper))
+      case _ =>
+        throw new Exception(notSupportedEngineError)
+    }
   }
 
   addReader(readModel)
@@ -437,6 +461,11 @@ trait ReadBertForZeroShotDLModel extends ReadTensorflowModel {
           .setSignatures(_signatures)
           .setModelIfNotSet(spark, Some(wrapper), None)
 
+      case ONNX.name =>
+        val onnxWrapper = OnnxWrapper.read(localModelPath, zipped = false, useBundle = true)
+        annotatorModel
+          .setModelIfNotSet(spark, None, Some(onnxWrapper))
+
       case _ =>
         throw new Exception(notSupportedEngineError)
     }

src/test/scala/com/johnsnowlabs/nlp/annotators/classifier/dl/BertForZeroShotClassificationTestSpec.scala

@@ -33,7 +33,7 @@ class BertForZeroShotClassificationTestSpec extends AnyFlatSpec {
   val candidateLabels =
     Array("urgent", "mobile", "travel", "movie", "music", "sport", "weather", "technology")
 
-  "BertForSBertForZeroShotClassification" should "correctly load custom model with extracted signatures" taggedAs SlowTest in {
+  "BertForZeroShotClassification" should "correctly load custom model with extracted signatures" taggedAs SlowTest in {
 
     val ddd = Seq(
       "I have a problem with my iphone that needs to be resolved asap!!",