mt5-small-nlg-all-crosswoz

This model is a fine-tuned version of mt5-small on CrossWOZ both user and system utterances.

Refer to ConvLab-3 for model description and usage.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.001
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• optimizer: Adafactor
• lr_scheduler_type: linear
• num_epochs: 10.0

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0+cu102
• Datasets 2.3.2
• Tokenizers 0.12.1