README.md

metadata

language:
  - en
license: apache-2.0
tags:
  - roberta
  - classification
  - dialog state tracking
  - natural language understanding
  - uncertainty
  - conversational system
  - task-oriented dialog
datasets:
  - ConvLab/multiwoz21
metrics:
  - Joint Goal Accuracy
  - Slot F1
  - Joint Goal Expected Calibration Error
model-index:
  - name: setsumbt-dst-nlu-multiwoz21
    results:
      - task:
          type: classification
          name: dialog state tracking
        dataset:
          type: ConvLab/multiwoz21
          name: MultiWOZ21
          split: test
        metrics:
          - type: Joint Goal Accuracy
            value: 51.8
            name: JGA
          - type: Slot F1
            value: 91.1
            name: Slot F1
          - type: Joint Goal Expected Calibration Error
            value: 12.7
            name: JECE

SetSUMBT-dst-nlu-multiwoz21

This model is a fine-tuned version SetSUMBT of roberta-base on MultiWOZ2.1. This model is a combined DST and NLU model and is a distribution distilled version of a ensemble of 5 models. This model should be used to produce uncertainty estimates for the dialogue belief state.

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: 3
• eval_batch_size: 16
• seed: 0
• gradient_accumulation_steps: 1
• optimizer: AdamW
• loss: Ensemble Distribution Distillation Loss
• 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