TransfusionOptions
Options & Configuration · AiDotNet.VisionLanguage.Unified
Configuration options for Transfusion: combined autoregressive and diffusion loss in single transformer.
Inheritance: ModelOptions → GenerativeVLMOptions → UnifiedVisionOptions → TransfusionOptions
For Beginners
These options configure the Transfusion model. Default values follow the original paper settings.
Constructors
| Constructor | Summary |
|---|---|
TransfusionOptions() |
Initializes a new instance with default values. |
TransfusionOptions(TransfusionOptions other) |
Initializes a new instance by copying from another instance. |
Properties
| Property | Summary |
|---|---|
bool EnableDiffusionLoss { get; set; } |
Gets or sets whether to use diffusion loss for image generation. |
double DiffusionLossWeight { get; set; } |
Gets or sets lambda, the coefficient that balances the diffusion loss against the language-modelling loss in L = L_LM + lambda * L_DDPM. |
int DiffusionInferenceSteps { get; set; } |
Gets or sets how many denoising steps GenerateImage takes over the trained schedule. |
int DiffusionTimesteps { get; set; } |
Gets or sets the number of training timesteps of the diffusion process. |
int? BeginningOfImageTokenId { get; set; } |
Gets or sets the beginning-of-image marker token id that precedes an image’s patches. |
int? EndOfImageTokenId { get; set; } |
Gets or sets the end-of-image marker token id that follows an image’s patches. |