Class LoRAConfiguration

Namespace

AiDotNet.Models.Options

Assembly

AiDotNet.dll

LoRA configuration for parameter-efficient fine-tuning.

public class LoRAConfiguration

Inheritance

object

LoRAConfiguration

Inherited Members

object.Equals(object)

object.Equals(object, object)

object.GetHashCode()

object.GetType()

object.MemberwiseClone()

object.ReferenceEquals(object, object)

object.ToString()

Properties

Alpha

Gets or sets the LoRA alpha scaling factor.

public int Alpha { get; set; }

Property Value

int

Remarks

Default: 32 (alpha/rank = 2 is common)

BiasMode

Gets or sets whether to use bias terms in LoRA layers.

public string BiasMode { get; set; }

Property Value

string

Dropout

Gets or sets the LoRA dropout rate.

public double Dropout { get; set; }

Property Value

double

QuantizationBits

Gets or sets the quantization bits when UseQuantization is true.

public int QuantizationBits { get; set; }

Property Value

int

Rank

Gets or sets the LoRA rank (r).

public int Rank { get; set; }

Property Value

int

Remarks

Lower rank = fewer parameters but less expressiveness.

Default: 16 (good balance)

TargetModules

Gets or sets which modules to apply LoRA to.

public string[] TargetModules { get; set; }

Property Value

string[]

Remarks

Common choices: ["q_proj", "v_proj"] for attention-only, or ["q_proj", "k_proj", "v_proj", "o_proj"] for full attention.

UseQuantization

Gets or sets whether to use quantization (QLoRA).

public bool UseQuantization { get; set; }

Property Value

bool