PixtralLarge<T>
Models & Types · AiDotNet.VisionLanguage.InstructionTuned
Pixtral Large: Mistral’s 124B decoder + 1B vision encoder.
Inheritance: NeuralNetworkBase<T> → VisionLanguageModelBase<T> → PixtralLarge<T>
Implements: IInstructionTunedVLM<T>
For Beginners
Pixtral Large is the scaled-up version of Pixtral, increasing from 12B to 124B total parameters with a 1B vision encoder (up from 400M). It uses Mistral-Large as the language backbone, which provides much stronger reasoning, instruction following, and knowledge capabilities than the standard Mistral model. The larger vision encoder also captures more detailed visual information. This model targets server-side deployment where maximum quality is more important than model size, making it suitable for high-stakes applications like medical image analysis, detailed document understanding, and complex visual reasoning. Default values follow the original paper settings.
How It Works
Pixtral Large (Mistral, 2024) scales up the Pixtral architecture to 124B parameters with a 1B vision encoder. It uses the Mistral-Large language backbone for improved reasoning and instruction following with high-resolution image understanding.
References:
- Paper: “Pixtral Large” (Mistral, 2024)
Example
// Create a Pixtral Large model for frontier visual reasoning
// using 1B vision encoder with Mistral-Large 124B 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 PixtralLarge<double>(architecture, "pixtrallarge.onnx");
// Training mode with native layers
var trainModel = new PixtralLarge<double>(architecture, new PixtralLargeOptions());
Constructors
| Constructor | Summary |
|---|---|
PixtralLarge(NeuralNetworkArchitecture<T> architecture, PixtralLargeOptions options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer) |
|
PixtralLarge(NeuralNetworkArchitecture<T> architecture, string modelPath, PixtralLargeOptions 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 from an image using Pixtral Large’s variable-resolution architecture. |