QVQ72B<T>

Models & Types · AiDotNet.VisionLanguage.Reasoning

QVQ-72B: first open-source multimodal reasoning model from Qwen.

Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → QVQ72B<T>

Implements: IReasoningVLM<T>

For Beginners

QVQ-72B is a large open-source multimodal model with visual chain-of-thought reasoning capabilities. Default values follow the original paper settings.

How It Works

QVQ-72B (Qwen Team, 2024) is the first open-source multimodal reasoning model, built on the Qwen architecture with 72 billion parameters. It features visual chain-of-thought reasoning where the model generates detailed step-by-step analysis of visual content before producing answers, achieving strong performance on mathematical and scientific visual reasoning benchmarks.

References:

  • Paper: “QVQ: To See the World with Wisdom” (Qwen Team, 2024)

Example

// Create a QVQ-72B model for open-source multimodal reasoning
// based on the Qwen architecture with visual chain-of-thought
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 QVQ72B<double>(architecture, "qvq72b.onnx");
            
// Training mode with native layers
var trainModel = new QVQ72B<double>(architecture, new QVQ72BOptions());

Constructors

Constructor Summary
QVQ72B(NeuralNetworkArchitecture<T> architecture, QVQ72BOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer)
QVQ72B(NeuralNetworkArchitecture<T> architecture, string modelPath, QVQ72BOptions 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 QVQ-72B’s Qwen2-VL encoder with visual reasoning.
Tensor<T> ReasonWithChainOfThought(Tensor<T> image, string question) Generates multi-step reasoning using QVQ-72B’s RL-aligned chain-of-thought.