UDOPOptions
Options & Configuration · AiDotNet.Document.Options
Configuration options for the UDOP document model.
Inheritance: ModelOptions → NeuralNetworkOptions → ModelHyperparameterOptions → DocumentNeuralNetworkOptions → UDOPOptions
Constructors
| Constructor | Summary |
|---|---|
UDOPOptions() |
Initializes a new instance of the UDOPOptions class with the published UDOP-large configuration. |
Properties
| Property | Summary |
|---|---|
double LayerNormEpsilon { get; set; } |
RMSNorm epsilon (layer_norm_epsilon). |
double LearningRate { get; set; } |
Initial learning rate. |
int FeedForwardDim { get; set; } |
Inner width of each ReLU feed-forward block (T5 d_ff). |
int KeyValueDim { get; set; } |
Width of each attention head (T5 d_kv). |
int Max2DPositions { get; set; } |
Rows in each 2-D cell-embedding table (max_2d_position_embeddings). |
int MaxGenerationLength { get; set; } |
Largest number of tokens a prompted decode (question answering, layout analysis) generates before it stops. |
int NumLocationBins { get; set; } |
Number of layout location tokens, quantising a 0-1 coordinate into this many bins. |
int RelativeAttentionBuckets { get; set; } |
Number of relative-position buckets shared by every bias table. |
int RelativeAttentionMaxDistance { get; set; } |
Largest 1-D token distance with its own bucket range (T5 max_distance). |
int RelativeAttentionMaxDistance2D { get; set; } |
Largest horizontal or vertical layout distance (0-1 coordinates scaled by 100) before buckets saturate. |
Methods
| Method | Summary |
|---|---|
void Validate() |
Throws if a value this model requires has been left unset or is not positive. |