Intent and Slot Classifier Evaluation Thresholds
- Gardner, I.V. (Bella)
To ensure that your intent and slot classifier is functional, it’s important to meet specific accuracy thresholds. Remarkably, these thresholds can be achieved quickly using the given dataset, hyperparameters, and default model configuration we provide. With just 2 epochs of training and no hyperparameter tuning, extended training, or advanced techniques, you should be able to meet these minimum standards. Below are the key hyperparameters for model training:
Model Training Parameters:
--num_epochs: Number of training epochs. Default: 2.--batch_size: Batch size for training. Default: 16.--learning_rate: Learning rate for the optimizer. Default: 5e-5.--max_length: Maximum sequence length for tokenization. Default: 16.
Miscellaneous:
--seed: Random seed for reproducibility. Default: 42.
Using these default settings, you should be able to build a functional model without additional fine-tuning or adjustments. This provides a strong baseline for your experiments and further optimization.
Below, you’ll find tables outlining the evaluation statistics you need to meet. These benchmarks are achievable using the default settings described above, ensuring your model is functional without additional tuning or advanced modifications.
