SigLIPTextConditioner<T>

Models & Types · AiDotNet.Diffusion.Conditioning

SigLIP text encoder conditioning module (Zhai et al., ICCV 2023). Same encoder architecture as CLIP; the paper’s contribution is the sigmoid contrastive loss (vs CLIP’s softmax), which is upstream of this encoder body.

Inheritance: NeuralNetworkBase<T> → TextConditioningBase<T> → SigLIPTextConditioner<T>

Constructors

Constructor Summary
SigLIPTextConditioner(ITokenizer tokenizer, NeuralNetworkArchitecture<T> architecture, SigLIPTextConditionerOptions options)

Properties

Property Summary
bool ProducesPooledOutput { get; }

Methods

Method Summary
SigLIPTextConditioner<T> FromPretrained(SigLIPVariant variant, string huggingFaceModelName, string cacheDir) Loads a paper-canonical SigLIP text conditioner with its real pretrained tokenizer from HuggingFace.