DistillationHyperParameters

@Serializable
data class DistillationHyperParameters(val adapterSize: AdapterSize? = null, val epochCount: Long? = null, val learningRateMultiplier: Double? = null, val generationConfig: GenerationConfig? = null, val learningRate: Double? = null, val batchSize: Int? = null)

Distillation hyperparameters for tuning.

Constructors

Link copied to clipboard
constructor(adapterSize: AdapterSize? = null, epochCount: Long? = null, learningRateMultiplier: Double? = null, generationConfig: GenerationConfig? = null, learningRate: Double? = null, batchSize: Int? = null)

Properties

Link copied to clipboard

The size of the adapter. Can be 'small', 'medium', or 'large'.

Link copied to clipboard
val batchSize: Int? = null

Batch size for tuning. This feature is only available for open source models.

Link copied to clipboard
val epochCount: Long? = null

Number of complete passes the model makes over the entire training dataset during training.

Link copied to clipboard

Generation config for Distillation teacher model sampling. Only the following fields are supported for distillation teacher samplings:

Link copied to clipboard
val learningRate: Double? = null

The learning rate for distillation tuning.

Link copied to clipboard

Multiplier for adjusting the default learning rate.