README.md

metadata

language:
  - en
license: apache-2.0
tags:
  - t5-small
  - text2text-generation
  - dialog state tracking
  - conversational system
  - task-oriented dialog
datasets:
  - ConvLab/multiwoz21
  - ConvLab/sgd
  - ConvLab/tm1
  - ConvLab/tm2
  - ConvLab/tm3
metrics:
  - Joint Goal Accuracy
  - Slot F1
model-index:
  - name: t5-small-dst-multiwoz21_sgd_tm1_tm2_tm3
    results:
      - task:
          type: text2text-generation
          name: dialog state tracking
        dataset:
          type: ConvLab/multiwoz21
          name: MultiWOZ 2.1
          split: test
          revision: 5f55375edbfe0270c20bcf770751ad982c0e6614
        metrics:
          - type: Joint Goal Accuracy
            value: 53.1
            name: JGA
          - type: Slot F1
            value: 91.9
            name: Slot F1
      - task:
          type: text2text-generation
          name: dialog state tracking
        dataset:
          type: ConvLab/sgd
          name: SGD
          split: test
          revision: 6e8c79b888b21cc658cf9c0ce128d263241cf70f
        metrics:
          - type: Joint Goal Accuracy
            value: 20.6
            name: JGA
          - type: Slot F1
            value: 60
            name: Slot F1
      - task:
          type: text2text-generation
          name: dialog state tracking
        dataset:
          type: ConvLab/tm1, ConvLab/tm2, ConvLab/tm3
          name: TM1+TM2+TM3
          split: test
        metrics:
          - type: Joint Goal Accuracy
            value: 48.6
            name: JGA
          - type: Slot F1
            value: 81
            name: Slot F1
widget:
  - text: >-
      multiwoz21: user: I would like a taxi from Saint John's college to Pizza
      Hut Fen Ditton.

      system: What time do you want to leave and what time do you want to arrive
      by?

      user: I want to leave after 17:15.
    example_title: MultiWOZ 2.1
  - text: >-
      sgd: user: Hi, could you get me a restaurant booking on the 8th please?

      system: Any preference on the restaurant, location and time?

      user: Could you get me a reservation at P.f. Chang's in Corte Madera at
      afternoon 12?
    example_title: Schema-Guided Dialog
  - text: |-
      tm1: user: Hi there, could you please help me with an order of Pizza?
      system: Sure, where would you like to order you pizza from?
      user: I would like to order a pizza from Domino's.
    example_title: Taskmaster-1
  - text: >-
      tm2: user: I need help finding a hotel in New Orleans.

      system: Okay.

      user: I need something that's around $300 a night and it's a five star
      rating.
    example_title: Taskmaster-2
  - text: |-
      tm3: user: Hi, I'm hoping to see a movie tonight.
      system: Great, I can assist with that. What genre of film do you prefer.
      user: I usually like comedies.
    example_title: Taskmaster-3
inference:
  parameters:
    max_length: 100

t5-small-dst-multiwoz21_sgd_tm1_tm2_tm3

This model is a fine-tuned version of t5-small on MultiWOZ 2.1, Schema-Guided Dialog, Taskmaster-1, Taskmaster-2, and Taskmaster-3.

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: 64
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• optimizer: Adafactor
• lr_scheduler_type: linear
• num_epochs: 10.0

Framework versions

• Transformers 4.20.1
• Pytorch 1.11.0+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1