metadata
language:
- en
license: apache-2.0
tags:
- t5-small
- text2text-generation
- natural language understanding
- conversational system
- task-oriented dialog
datasets:
- ConvLab/sgd
metrics:
- Dialog acts Accuracy
- Dialog acts F1
model-index:
- name: t5-small-nlu-sgd
results:
- task:
type: text2text-generation
name: natural language understanding
dataset:
type: ConvLab/sgd
name: SGD
split: test
revision: 6e8c79b888b21cc658cf9c0ce128d263241cf70f
metrics:
- type: Dialog acts Accuracy
value: 45
name: Accuracy
- type: Dialog acts F1
value: 58.6
name: F1
widget:
- text: >-
user: Could you get me a reservation at P.f. Chang's in Corte Madera at
afternoon 12?
- text: >-
user: Sure, may I know if they have vegetarian options and how expensive
is their food?
inference:
parameters:
max_length: 100
t5-small-nlu-sgd
This model is a fine-tuned version of t5-small on Schema-Guided Dialog.
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: 128
- eval_batch_size: 64
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 256
- optimizer: Adafactor
- lr_scheduler_type: linear
- num_epochs: 10.0
Framework versions
- Transformers 4.18.0
- Pytorch 1.10.2+cu102
- Datasets 1.18.3
- Tokenizers 0.11.0