KimiVLThinking<T>

Models & Types · AiDotNet.VisionLanguage.Reasoning

Kimi-VL-Thinking: long chain-of-thought reasoning with RL alignment.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → KimiVLThinking<T>

Implements: IReasoningVLM<T>

For Beginners

Kimi-VL-Thinking is a vision-language model specialized in long chain-of-thought reasoning for complex visual tasks. Default values follow the original paper settings.

How It Works

Kimi-VL-Thinking (Moonshot AI, 2025) extends Kimi-VL with long chain-of-thought reasoning capabilities trained through reinforcement learning alignment. It generates structured multi-step reasoning traces for complex visual problems, breaking down tasks into intermediate steps before producing final answers.

References:

  • Paper: “Kimi-VL Technical Report” (Moonshot AI, 2025)

Example

// Create a Kimi-VL-Thinking model for long chain-of-thought visual reasoning
// with RL alignment for structured multi-step problem solving
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.TwoDimensional,
    taskType: NeuralNetworkTaskType.ImageClassification,
    inputHeight: 224, inputWidth: 224, inputDepth: 3, outputSize: 512);
            
// ONNX inference mode with pre-trained model
var model = new KimiVLThinking<double>(architecture, "kimivlthinking.onnx");
            
// Training mode with native layers
var trainModel = new KimiVLThinking<double>(architecture, new KimiVLThinkingOptions());

Constructors

Constructor Summary
KimiVLThinking(NeuralNetworkArchitecture<T> architecture, KimiVLThinkingOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
KimiVLThinking(NeuralNetworkArchitecture<T> architecture, string modelPath, KimiVLThinkingOptions options)

Properties

Property Summary
int DecoderEmbeddingDim { get; }
int EmbeddingDimension { get; }
int MaxGenerationLength { get; }
string LanguageModelName { get; }
string ReasoningApproach { get; }

Methods

Method Summary
ModelMetadata<T> GetModelMetadata()
Tensor<T> Chat(Tensor<T> image, IEnumerable<ValueTuple<string, string>> conversationHistory, string userMessage)
Tensor<T> EncodeImage(Tensor<T> image)
Tensor<T> GenerateFromImage(Tensor<T> image, string prompt) Generates text from image using Kimi-VL-Thinking’s RL-aligned long thinking pipeline.
Tensor<T> ReasonWithChainOfThought(Tensor<T> image, string question) Generates extended reasoning using Kimi-VL-Thinking’s RL-aligned long thinking chains.