I have a problem with running google/mobilebert-uncased in Xamarin.
I get an error System.DllNotFoundException Message=onnxruntime assembly:<unknown assembly> type:<unknown type> member:(null) in the line session = new InferenceSession(modelPath);.
I am using Visual Studio 2022 and .NET Standard 2.1. My Microsoft.ML.OnnxRuntime.dll file is located at objDebug110androidassets.
And I took all the files from google/mobilebert-uncased on the official HuggingFace site (except I compiled the model from .bin to .onnx).

My code:

using System.Collections.Generic;
using FastBertTokenizer;
using Microsoft.ML.Data;
using Microsoft.ML.OnnxRuntime;
using Microsoft.ML.OnnxRuntime.Tensors;
using Xamarin.Forms;
using System.Linq;
using System.Threading.Tasks;

namespace manipulatorMobileApp.Bert
{
    public class BertInput
    {
        [ColumnName("input_ids")]
        public long[] InputIds { get; set; }

        [ColumnName("attention_mask")]
        public long[] AttentionMask { get; set; }
    }
    public class MobileBert
    {
        public static async Task<string> PredictWord(string text)
        {
            InferenceSession session = null;
            try
            {
                BertTokenizer tokenizer = new BertTokenizer();
                string tokenizerFilePath = DependencyService.Get<IFileHelper>().GetFilePath("tokenizer.json");
                await tokenizer.LoadTokenizerJsonAsync(tokenizerFilePath);
                var (inputIds, attentionMask, tokenTypeIds) = tokenizer.Encode(text);

                BertInput bertInput = new BertInput()
                {
                    InputIds = inputIds.ToArray(),
                    AttentionMask = attentionMask.ToArray(),
                };

                string modelPath = DependencyService.Get<IFileHelper>().GetFilePath("mobilebert.onnx");

                DenseTensor<long> inputTensor = new DenseTensor<long>(bertInput.InputIds,
                    new[] { 1, bertInput.InputIds.Length });
                DenseTensor<long> attentionTensor = new DenseTensor<long>(bertInput.AttentionMask,
                    new[] { 1, bertInput.AttentionMask.Length });

                List<NamedOnnxValue> input = new List<NamedOnnxValue>
                {
                    NamedOnnxValue.CreateFromTensor("input_ids", inputTensor),
                    NamedOnnxValue.CreateFromTensor("attention_mask", attentionTensor)
                };

                session = new InferenceSession(modelPath);
                var output = session.Run(input);

                var prediction = output.First().AsEnumerable<long>().ToArray();
                var decoded = tokenizer.Decode(prediction);
                var predictedWord = decoded.Split(' ').FirstOrDefault();

                return predictedWord;
            }
            finally
            {
                if (session != null)
                {
                    session.Dispose();
                }
            }
        }
    }
}

Libraries I use:

If you need more information, I’m happy to provide it to you.

I’ve tried re-converting the model to .onnx, I’ve also reinstalled the libraries and tried different combinations of them. I also downloaded the latest versions of Visual C++ Redistributable.

I have to ask for your help as I have not found an answer to this question on other forums. Thank you and feel free to ask for more information if you also decide to look into this issue!