|
--- |
|
license: mit |
|
datasets: |
|
- kasnerz/logic2text |
|
language: |
|
- en |
|
--- |
|
|
|
## Model description |
|
|
|
This is a fine-tuned [t5-base](https://huggingface.co/t5-base) model for generating logical form templates from table metadata. |
|
It is a part of table-to-text pipeline with content selection and logical reasoning, trained on [Logic2Text dataset](https://aclanthology.org/2020.findings-emnlp.190/). |
|
The last pipeline step is done by [kategaranina/lt2t_lf_to_text](https://huggingface.co/kategaranina/lt2t_lf_to_text) model. |
|
For relevant code, see [LT2T repository](https://github.com/kategerasimenko/LT2T) on GitHub. |
|
|
|
Example of input: |
|
|
|
``` |
|
Title: wake forest demon deacons football , 1980 - 89. |
|
Column date. Type: date. Stats: Column has all unique values, all values have one occurrence, |
|
first: 1987-09-12, last: 1987-11-21, max diff days: 70, max diff months: 2, max diff years: 0. |
|
Value: 09 / 12 / 1987. |
|
... |
|
Column location. Type: str. Stats: Column has most unique values, one value has most occurrences, |
|
most values have one occurrence. Value: groves stadium winston - salem , nc. |
|
... |
|
``` |
|
|
|
Example of output: |
|
``` |
|
round_eq { avg { all_rows ; attendance } ; X } |
|
``` |
