README.md

metadata

license: apache-2.0
datasets:
  - wikimedia/wikipedia
  - SiberiaSoft/SiberianPersonaChat-2
language:
  - ru
  - en
metrics:
  - mse
tags:
  - AIFramework 2.2 Open
  - .net
  - sbert
  - C#

Porting this model for use in C#

• Example of use in C#:

  using AI.DataPrepaire.DataLoader.NNWBlockLoader;
  using AI.DataStructs.Algebraic;
  using AI.ONNX.NLP.Bert;
  
  
  string path = "SbertDistilAIFr";
  // Загрузка модели Bert
  BertEmbedder embedder = BertEmbedder.FromPretrained(path);
  // Добавление последнего(линейного) слоя
  LinearLayerLoader linearLayer = LinearLayerLoader.LoadFromBinary(@$"{path}\1_Linear\model.aifr");
  embedder.V2VBlocks.Add(linearLayer);
  
  // Векторизация предложений
  Vector vects1 = embedder.ForwardSBert("Kittens love milk");
  Vector vects2 = embedder.ForwardSBert("Visual Studio Code (VS Code) — текстовый редактор, разработанный Microsoft для Windows, Linux и macOS.");
  Vector vects3 = embedder.ForwardSBert("Visual Studio Code, also commonly referred to as VS Code,[12] is a source-code editor developed by Microsoft for Windows, Linux and macOS.");
  Vector vects1_rus = embedder.ForwardSBert("Котята любят молоко");
  
  // Рассчет близости между текстами
  Console.WriteLine(vects1.Cos(vects1_rus));
  Console.WriteLine(vects1.Cos(vects2));
  Console.WriteLine(vects2.Cos(vects3));
  Console.WriteLine(vects1.Cos(vects3));
  Console.ReadKey();
  

• Network output

0.8312274467817271
0.18402859300814606
0.9556528309990797
0.14738954582438052

• GitHub