PretrainedModelLoader<T>
Models & Types · AiDotNet.ModelLoading
Loads pretrained models from various sources.
For Beginners
This is your gateway to using pretrained models.
Instead of training models from scratch (which requires massive datasets and compute resources), you can load pretrained weights that others have trained.
Example usage:
var loader = new PretrainedModelLoader<float>();
// Load a pretrained VAE encoder
var encoder = new VAEEncoder<float>(inputChannels: 3, latentChannels: 4);
var loadResult = loader.LoadWeights(encoder, "sd-vae-ft-mse/diffusion_pytorch_model.safetensors");
// Now your VAE is ready for image encoding/decoding!
How It Works
This class provides methods to load pretrained model weights into our model classes. It supports loading from local SafeTensors files and handles weight name mapping between different model formats.
Constructors
| Constructor | Summary |
|---|---|
PretrainedModelLoader(bool verbose) |
Initializes a new instance of the PretrainedModelLoader class. |
Methods
| Method | Summary |
|---|---|
Dictionary<string, Tensor<T>> LoadAllTensors(string path) |
Loads all tensors from a SafeTensors file. |
Dictionary<string, Tensor<T>> LoadTensors(string path, IEnumerable<string> tensorNames) |
Loads specific tensors by name from a SafeTensors file. |
List<TensorMetadata> GetTensorInfo(string path) |
Gets information about tensors in a SafeTensors file. |
LoadResult LoadWeights(IWeightLoadable<T> model, string weightsPath, Func<string, string> mapping, bool strict) |
Loads weights using a WeightMapping instance. |
LoadResult LoadWeights(IWeightLoadable<T> model, string weightsPath, WeightMapping mapping, bool strict) |
Loads weights using a WeightMapping instance. |
ValidationResult ValidateWeights(string path, IEnumerable<string> requiredTensorPatterns) |
Validates that required tensors exist in a weights file. |