DistillationCheckpointConfig
Options & Configuration · AiDotNet.KnowledgeDistillation
Configuration for distillation checkpoint management.
For Beginners
This class controls when and how models are saved during knowledge distillation training. Think of it like the “auto-save” settings in a video game.
Constructors
| Constructor | Summary |
|---|---|
DistillationCheckpointConfig() |
Properties
| Property | Summary |
|---|---|
bool LowerIsBetter { get; set; } |
Whether lower metric values are better (true for loss, false for accuracy). |
bool SaveCurriculumState { get; set; } |
Save curriculum strategy state with checkpoint. |
bool SaveMetadata { get; set; } |
Save training metadata (epoch number, loss history, etc.). |
bool SaveStudent { get; set; } |
Save the student model checkpoint. |
bool SaveTeacher { get; set; } |
Save the teacher model checkpoint. |
int KeepBestN { get; set; } |
Keep only the best N checkpoints based on validation metric. |
int SaveEveryBatches { get; set; } |
Save checkpoint every N batches (0 = disabled). |
int SaveEveryEpochs { get; set; } |
Save checkpoint every N epochs (0 = disabled). |
string BestMetric { get; set; } |
Metric to use for determining “best” checkpoint (e.g., “validation_loss”, “accuracy”). |
string CheckpointDirectory { get; set; } |
Directory where checkpoints will be saved. |
string CheckpointPrefix { get; set; } |
Prefix for checkpoint filenames. |