adenhaus commited on
Commit
186968a
1 Parent(s): 38c3ecc

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +5 -2
README.md CHANGED
@@ -18,13 +18,16 @@ multilinguality:
18
  license: cc-by-sa-4.0
19
  inference: false
20
  ---
 
21
 
22
  This learned metric is for evaluating models trained on the TaTA dataset. It was trained as per instructions in [TaTA: A Multilingual Table-to-Text Dataset for African Languages](https://aclanthology.org/2023.findings-emnlp.118/) (StATA-QE variant).
23
 
24
- It achieves an RMSE loss of 0.41 on the dev split, and a Pearson correlation of 0.23 with human evaluations on the test split ("attributable" column of [this dataset](https://huggingface.co/datasets/adenhaus/stata)).
25
-
26
  The original file can be found [here](https://github.com/google-research/url-nlp/tree/main/tata).
27
 
 
 
 
 
28
  [Here](https://huggingface.co/adenhaus/mt5-large-stata/blob/main/README.md) is a version trained with mT5-Large instead of mT5-Small, which achieves a correlation of 0.59.
29
 
30
  # Example use
 
18
  license: cc-by-sa-4.0
19
  inference: false
20
  ---
21
+ # Background
22
 
23
  This learned metric is for evaluating models trained on the TaTA dataset. It was trained as per instructions in [TaTA: A Multilingual Table-to-Text Dataset for African Languages](https://aclanthology.org/2023.findings-emnlp.118/) (StATA-QE variant).
24
 
 
 
25
  The original file can be found [here](https://github.com/google-research/url-nlp/tree/main/tata).
26
 
27
+ # Performance
28
+
29
+ It achieves an RMSE loss of 0.41 on the dev split, and a Pearson correlation of 0.23 with human evaluations on the test split ("attributable" column of [this dataset](https://huggingface.co/datasets/adenhaus/stata)).
30
+
31
  [Here](https://huggingface.co/adenhaus/mt5-large-stata/blob/main/README.md) is a version trained with mT5-Large instead of mT5-Small, which achieves a correlation of 0.59.
32
 
33
  # Example use