OpenCLIPOptions

Options & Configuration · AiDotNet.VisionLanguage.Encoders

Configuration options for the OpenCLIP model.

Inheritance: ModelOptions → ContrastiveEncoderOptions → OpenCLIPOptions

For Beginners

OpenCLIP is essentially the same as CLIP but trained on publicly available data instead of proprietary data. This makes it more transparent and reproducible while achieving similar or better performance. It supports more model sizes and configurations.

How It Works

OpenCLIP (Ilharco et al., 2021) is an open-source reproduction of CLIP trained on the LAION-2B and LAION-5B datasets. It supports a wide range of vision encoder architectures (ViT-B/32 through ViT-bigG/14) and achieves comparable or better performance than OpenAI’s original CLIP.

Constructors

Constructor Summary
OpenCLIPOptions() Initializes a new instance with default values.
OpenCLIPOptions(OpenCLIPOptions other) Initializes a new instance by copying from another instance.

Properties

Property Summary
ContrastiveLossType LossType { get; set; } Gets or sets the contrastive loss type.
ModelPrecision Precision { get; set; } Gets or sets the precision mode for inference.
PretrainingDataset Dataset { get; set; } Gets or sets the pre-training dataset.
bool UseCoCaVariant { get; set; } Gets or sets whether to use the CoCa (Contrastive Captioners) variant.