Update to OnnxRuntime 1.6.0 and fixed bug with sequences outputs (#5529)

* Use onnx prerelease * Upgrade to onnx 1.6.0 * Updated docs * Fixed problem with sequences

Author: antoniovs1029

eng/Versions.props

@@ -23,7 +23,7 @@
  <GoogleProtobufPackageVersion>3.10.1</GoogleProtobufPackageVersion>
  <LightGBMPackageVersion>2.2.3</LightGBMPackageVersion>
  <MicrosoftExtensionsPackageVersion>2.1.0</MicrosoftExtensionsPackageVersion>
- <MicrosoftMLOnnxRuntimePackageVersion>1.5.2</MicrosoftMLOnnxRuntimePackageVersion>
+ <MicrosoftMLOnnxRuntimePackageVersion>1.6.0</MicrosoftMLOnnxRuntimePackageVersion>
  <MlNetMklDepsPackageVersion>0.0.0.9</MlNetMklDepsPackageVersion>
  <ParquetDotNetPackageVersion>2.1.3</ParquetDotNetPackageVersion>
  <SystemDrawingCommonPackageVersion>4.5.0</SystemDrawingCommonPackageVersion>

src/Microsoft.ML.OnnxTransformer/OnnxTransform.cs

@@ -768,7 +768,7 @@ public NamedOnnxValue GetNamedOnnxValue()
  /// | Does this estimator need to look at the data to train its parameters? | No |
  /// | Input column data type | Known-sized vector of <xref:System.Single> or <xref:System.Double> types |
  /// | Output column data type | As specified by the ONNX model |
- /// | Required NuGet in addition to Microsoft.ML | Microsoft.ML.OnnxTransformer (always), either Microsoft.ML.OnnxRuntime 1.5.2 (for CPU processing) or Microsoft.ML.OnnxRuntime.Gpu 1.5.2 (for GPU processing if GPU is available) |
+ /// | Required NuGet in addition to Microsoft.ML | Microsoft.ML.OnnxTransformer (always), either Microsoft.ML.OnnxRuntime 1.6.0 (for CPU processing) or Microsoft.ML.OnnxRuntime.Gpu 1.6.0 (for GPU processing if GPU is available) |
  /// | Exportable to ONNX | No |
  ///
  /// To create this estimator use the following APIs:

src/Microsoft.ML.OnnxTransformer/OnnxTypeParser.cs

@@ -267,7 +267,13 @@ private class CastHelper
 
  public static IEnumerable<TDst> CastOnnxSequenceToIEnumerable<TSrc, TDst>(IEnumerable<TSrc> o, Func<TSrc, object> caster)
  {
- return o.Select(v => (TDst)caster(v));
+ // Since now we're disposing the NamedOnnxValue objects
+ // after running inference on each output, we need
+ // to copy (enumerate) the output through ".ToList()"
+ // else, if our users try the keep the past sequence
+ // outputs of their OnnxTransformer, they would
+ // end up with empty sequences.
+ return o.Select(v => (TDst)caster(v)).ToList();
  }
  }