OFASubnetSummary
Models & Types · AiDotNet.Models.Results
Represents OnceForAll subnet configuration summary.
For Beginners
OnceForAll (OFA) is a NAS technique that trains one large “supernet” containing many possible architectures. After training, you can extract smaller “subnets” optimized for specific hardware without retraining. This summary describes the extracted subnet.
Constructors
| Constructor | Summary |
|---|---|
OFASubnetSummary() |
Properties
| Property | Summary |
|---|---|
List<double> ExpansionRatios { get; set; } |
Gets or sets the expansion ratios per block. |
List<double> Widths { get; set; } |
Gets or sets the width multipliers per stage. |
List<int> Depths { get; set; } |
Gets or sets the depth configuration (number of blocks per stage). |
List<int> KernelSizes { get; set; } |
Gets or sets the kernel sizes selected per block. |
double? EstimatedAccuracy { get; set; } |
Gets or sets the estimated accuracy of the subnet. |
double? EstimatedLatencyMs { get; set; } |
Gets or sets the estimated latency of the subnet on the target platform. |