README.md

metadata

language:
  - zh
license: apache-2.0
tags:
  - mt5-small
  - text2text-generation
  - natural language generation
  - conversational system
  - task-oriented dialog
datasets:
  - ConvLab/crosswoz
metrics:
  - Slot Error Rate
  - sacrebleu
model-index:
  - name: mt5-small-nlg-all-crosswoz
    results:
      - task:
          type: text2text-generation
          name: natural language generation
        dataset:
          type: ConvLab/crosswoz
          name: CrossWOZ
          split: test
          revision: 4a3e56082543ed9eecb9c76ef5eadc1aa0cc5ca0
        metrics:
          - type: Slot Error Rate
            value: 6.9
            name: SER
          - type: sacrebleu
            value: 21
            name: BLEU
widget:
  - text: >-
      [Inform][酒店]([价格][100-200元],[评分][5分]);[greet][General]([][]);[Request][酒店]([名称][])


      user: 
  - text: >-
      [Recommend][酒店]([名称][北京京仪大酒店],[名称][北京贵都大酒店]);[Inform][酒店]([酒店设施-健身房-否][]);[NoOffer][酒店]([][])


      system: 
inference:
  parameters:
    max_length: 100

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