InternVL3<T>
Models & Types · AiDotNet.VisionLanguage.InstructionTuned
InternVL3: 78B SOTA open-source model with advanced training and test-time recipes.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → InternVL3<T>
Implements: IInstructionTunedVLM<T>
For Beginners
InternVL3 is the largest and most capable model in the InternVL series, with 78 billion total parameters. It uses InternViT-6B as the vision encoder and InternLM3 as the language backbone with 8192 hidden dimensions and 80 decoder layers. It achieves state-of-the-art performance among open-source models on challenging benchmarks like MMMU (72.2 score). The model uses advanced training recipes and test-time strategies (like chain-of-thought prompting) to push open-source multimodal models to new performance levels. Default values follow the original paper settings.
How It Works
InternVL3 (2025) is the largest open-source vision-language model with 78B parameters, achieving 72.2 on MMMU (SOTA among open-source models). It uses InternViT-6B vision encoder with InternLM3 language backbone at 8192 hidden dimension and 80 decoder layers.
References:
- Paper: “InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models” (2025)
Example
// Create an InternVL3 model for state-of-the-art multimodal understanding
// using InternViT-6B with InternLM3 78B backbone
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 InternVL3<double>(architecture, "internvl3.onnx");
// Training mode with native layers
var trainModel = new InternVL3<double>(architecture, new InternVL3Options());
Constructors
| Constructor | Summary |
|---|---|
InternVL3(NeuralNetworkArchitecture<T> architecture, InternVL3Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
InternVL3(NeuralNetworkArchitecture<T> architecture, string modelPath, InternVL3Options options) |
Properties
| Property | Summary |
|---|---|
int DecoderEmbeddingDim { get; } |
|
int EmbeddingDimension { get; } |
|
int MaxGenerationLength { get; } |
|
string LanguageModelName { 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 using InternVL3’s advanced training and test-time recipes. |