PromptNEROptions
Options & Configuration · AiDotNet.NER.Options
Configuration for PromptNER’s paper architecture and training recipe.
Inheritance: ModelOptions → NeuralNetworkOptions → TransformerNEROptions → PromptNEROptions
How It Works
The reference configuration uses BERT-large, Adam with a peak learning rate of
2e-5, and a linear warmup followed by linear decay over 50-100 epochs.
The step counts remain public because the exact number of optimizer steps depends
on the caller’s dataset and batch size.
Constructors
| Constructor | Summary |
|---|---|
PromptNEROptions() |
Creates the reference-paper configuration. |
PromptNEROptions(PromptNEROptions other) |
Creates an independent copy of another PromptNER configuration. |
Properties
| Property | Summary |
|---|---|
bool EnableGradientClipping { get; set; } |
Gets or sets whether global-norm gradient clipping is enabled. |
double AdamBeta1 { get; set; } |
Gets or sets Adam’s first-moment coefficient. |
double AdamBeta2 { get; set; } |
Gets or sets Adam’s second-moment coefficient. |
double AdamEpsilon { get; set; } |
Gets or sets Adam’s numerical-stability epsilon. |
double MaxGradientNorm { get; set; } |
Gets or sets the global gradient-norm limit when clipping is enabled. |