System.InvalidOperationException: Variable length input columns not supported on multidimensional array

System Information (please complete the following information):

Windows 11
ML.NET 1.7.1
.NET 6.0

Describe the bug System.InvalidOperationException: Variable length input columns not supported on when using VectorTypeAttribute or shapeDictionary parameter on OnnxCatalog.ApplyOnnxModel

To Reproduce Steps to reproduce the behavior:

Create new Console Project
Add Microsoft.ML.OnnxTransformer

[VectorType(0,1, 1, 16)] or

var onnxPredictionPipeline = mlContext.Transforms.ApplyOnnxModel(outputColumns, inputColumns,
 "model.onnx", shapeDictionary: new Dictionary<string, int[]>
{
    ["context_word"] = new int[] { 0, 1 },
    ["context_char"] = new int[] { 0, 1, 1, 16 },
    ["query_word"] = new int[] { 0, 1 },
["query_char"] = new int[] { 0, 1, 1, 16 },
},
 null, false, 100);

will cause the error when TrivialEstimator.Fit() is called

Screenshots, Code, Sample Projects I'm trying to use this Onnx model https://github.com/onnx/models/tree/main/text/machine_comprehension/bidirectional_attention_flow

context_word: [seq, 1,] of string
context_char: [seq, 1, 1, 16] of string
query_word: [seq, 1,] of string
query_char: [seq, 1, 1, 16] of string

Full Code

using Microsoft.ML.Transforms.Onnx;
using Microsoft.ML;

namespace MachineLearning
{
    internal class Program
    {
        static void Main(string[] args)
        {
            MLContext mlContext = new MLContext();
            var onnxPredictionPipeline = GetPredictionPipeline(mlContext);
            var onnxPredictionEngine = mlContext.Model.CreatePredictionEngine<OnnxInput, OnnxOutput>(onnxPredictionPipeline);
            var testInput = new OnnxInput("A quick brown fox jumps over the lazy dog.", "What color is the fox?");

            var testOutput = onnxPredictionEngine.Predict(testInput);
            Console.WriteLine($"start:{testOutput.StartPos} end:{testOutput.EndPos} ");
        }

        static ITransformer GetPredictionPipeline(MLContext mlContext)
        {
            var inputColumns = new string[]
            {
                "context_word", "query_word","context_char","query_char"
            };

            var outputColumns = new string[] { "start_pos", "end_pos" };
            var onnxPredictionPipeline = mlContext.Transforms.ApplyOnnxModel(outputColumns, inputColumns,
                    "model.onnx", shapeDictionary: new Dictionary<string, int[]>
                    {
                        ["context_word"] = new int[] { 0, 1 },
                        ["context_char"] = new int[] { 0, 1, 1, 16 },
                        ["query_word"] = new int[] { 0, 1 },
                        ["query_char"] = new int[] { 0, 1, 1, 16 },
                    }, null, false, 100);
            var Dv = mlContext.Data.LoadFromEnumerable(new OnnxInput[] { new OnnxInput("A quick brown fox jumpes over the lazy dog.", "What color is the fox?") });
            return onnxPredictionPipeline.Fit(Dv);
        }
    }
}

Output model

using Microsoft.ML.Data;

namespace MachineLearning
{
    public class OnnxOutput
    {
        [ColumnName("start_pos")]
        public int StartPos { get; set; }
        [ColumnName("end_pos")]
        public int EndPos { get; set; }
    }
}

Input model

using Microsoft.ML.Data;

namespace MachineLearning
{
    public class OnnxInput
    {
        public OnnxInput(string context, string query)
        {
        }

        [ColumnName("context_word")]
        public string[] ContextWord { get; set; }

        [ColumnName("context_char")]
        [VectorType(0,1, 1, 16)]
        public string[] ContextChar { get; set; }

        [ColumnName("query_word")]
        //[VectorType(-1,1)]
        public string[] QueryWord { get; set; }

        [ColumnName("query_char")]
        //[VectorType(-1,1,1,16)]
        public string[] QueryChar { get; set; }

    }
}

dotnet / machinelearning

System.InvalidOperationException: Variable length input columns not supported on multidimensional array #6387