Eagle25<T>
Models & Types · AiDotNet.VisionLanguage.InstructionTuned
Eagle 2.5: extended context VLM with video understanding capabilities.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → Eagle25<T>
Implements: IInstructionTunedVLM<T>
For Beginners
Eagle 2.5 extends the original Eagle model with the ability to understand videos and long sequences of images. Through specialized long-context post-training, it can process many visual frames in sequence — useful for video understanding, multi-page document analysis, and tasks that require tracking information across many images. It maintains strong single-image performance while adding this extended context capability. Default values follow the original paper settings.
How It Works
Eagle 2.5 (NVIDIA, 2025) extends Eagle with long-context post-training to handle extended visual sequences including video understanding. It uses specialized post-training techniques to boost the model’s ability to process long sequences of visual frames while maintaining strong performance on single-image tasks. The model supports extended context windows for processing multi-frame video content and long document sequences.
References:
- Paper: “Eagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language Models” (2025)
Example
// Create an Eagle 2.5 model for long-context video understanding
// with extended visual sequence support
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 Eagle25<double>(architecture, "eagle25.onnx");
// Training mode with native layers
var trainModel = new Eagle25<double>(architecture, new Eagle25Options());
Constructors
| Constructor | Summary |
|---|---|
Eagle25(NeuralNetworkArchitecture<T> architecture, Eagle25Options options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
Eagle25(NeuralNetworkArchitecture<T> architecture, string modelPath, Eagle25Options 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 Eagle 2.5’s adaptive tiling with long-context support. |