Amazon Titan Text Premier model support the following hyperparameters for model customization:
| Hyperparameter (console) | Hyperparameter (API) | Definition | Type | Minimum | Maximum | Default |
|---|---|---|---|---|---|---|
| Epochs | epochCount | The number of iterations through the entire training dataset | integer | 1 | 5 | 2 |
| Batch size (micro) | batchSize | The number of samples processed before updating model parameters | integer | 1 | 1 | 1 |
| Learning rate | learningRate | The rate at which model parameters are updated after each batch | float | 1.00E-07 | 1.00E-05 | 1.00E-06 |
| Learning rate warmup steps | learningRateWarmupSteps | The number of iterations over which the learning rate is gradually increased to the specified rate | integer | 0 | 20 | 5 |
Amazon Titan Text models, such as Lite and Express, support the following hyperparameters for model customization:
| Hyperparameter (console) | Hyperparameter (API) | Definition | Type | Minimum | Maximum | Default |
|---|---|---|---|---|---|---|
| Epochs | epochCount | The number of iterations through the entire training dataset | integer | 1 | 10 | 5 |
| Batch size (micro) | batchSize | The number of samples processed before updating model parameters | integer | 1 | 64 | 1 |
| Learning rate | learningRate | The rate at which model parameters are updated after each batch | float | 0.0 | 1 | 1.00E-5 |
| Learning rate warmup steps | learningRateWarmupSteps | The number of iterations over which the learning rate is gradually increased to the specified rate | integer | 0 | 250 | 5 |
