LLaVAOneVision15<T>
Models & Types · AiDotNet.VisionLanguage.InstructionTuned
LLaVA-OneVision-1.5: fully open model outperforming Qwen2.5-VL on 18/27 benchmarks.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → LLaVAOneVision15<T>
Implements: IInstructionTunedVLM<T>
For Beginners
LLaVA-OneVision 1.5 is the fully open successor that upgrades the language backbone to Qwen2.5 and uses improved training data and strategies. It supports single images, multi-image comparisons, and long video understanding with up to 64 frames. Notably, it outperforms the much larger Qwen2.5-VL on 18 out of 27 benchmarks while being fully open-source with all training data, code, and model weights publicly available. This makes it one of the most capable and accessible open multimodal models. Default values follow the original paper settings.
How It Works
LLaVA-OneVision-1.5 (Li et al., 2025) is the fully open successor to LLaVA-OneVision, using Qwen2.5 as the language backbone with improved training data and strategy. It supports single images, multi-image, and long video understanding with up to 64 frames.
References:
- Paper: “LLaVA-OneVision 1.5: Improved and Fully Open” (Li et al., 2025)
Example
// Create a LLaVA-OneVision 1.5 model for fully open multimodal AI
// supporting images, multi-image, and video with Qwen2.5 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 LLaVAOneVision15<double>(architecture, "llavaonevision15.onnx");
// Training mode with native layers
var trainModel = new LLaVAOneVision15<double>(architecture, new LLaVAOneVision15Options());
Constructors
| Constructor | Summary |
|---|---|
LLaVAOneVision15(NeuralNetworkArchitecture<T> architecture, LLaVAOneVision15Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
LLaVAOneVision15(NeuralNetworkArchitecture<T> architecture, string modelPath, LLaVAOneVision15Options 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 LLaVA-OneVision-1.5’s improved unified architecture. |