SetSUMBT-dst-tm1-tm2-tm3
This model is a fine-tuned version SetSUMBT of roberta-base on Taskmaster1, Taskmaster2 and Taskmaster3.
Refer to ConvLab-3 for model description and usage.
Training procedure
Training hyperparameters
The following hyperparameters were used during training:
- learning_rate: 0.00001
- train_batch_size: 2
- eval_batch_size: 8
- seed: 0
- gradient_accumulation_steps: 1
- optimizer: AdamW
- lr_scheduler_type: linear
- num_epochs: 50.0
Framework versions
- Transformers 4.17.0
- Pytorch 1.8.0+cu110
- Datasets 2.3.2
- Tokenizers 0.12.1
- Downloads last month
- 5
Dataset used to train ConvLab/setsumbt-dst-tm123
Evaluation results
- JGA on TM1+TM2+TM3test set self-reported24.900
- Slot F1 on TM1+TM2+TM3test set self-reported65.500