mrchtr commited on
Commit
5f95c71
1 Parent(s): fc9044a

Update model card

Browse files
Files changed (1) hide show
  1. README.md +32 -0
README.md CHANGED
@@ -1,3 +1,35 @@
1
  ---
 
 
 
 
 
 
 
 
2
  license: apache-2.0
 
 
3
  ---
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
  ---
2
+ language:
3
+ - de
4
+ tags:
5
+ - pytorch
6
+ - query-generation
7
+ widget:
8
+ - text: "Das Lama (Lama glama) ist eine Art der Kamele. Es ist in den südamerikanischen Anden verbreitet und eine vom Guanako abstammende Haustierform."
9
+ example_title: "Article 1"
10
  license: apache-2.0
11
+ metrics:
12
+ - Rouge-Score
13
  ---
14
+ # mt5-small-german-query-generation
15
+
16
+ ## Model description:
17
+ This model was created with the purpose to generate possible queries for a german input article.
18
+
19
+ For this model, we finetuned a multilingual T5 model [mt5-small](https://huggingface.co/google/mt5-small) on the [MMARCO dataset](https://huggingface.co/datasets/unicamp-dl/mmarco) the machine translated version of the MS MARCO dataset.
20
+
21
+
22
+ The model was trained for 1 epoch, on 200,000 unique queries of the dataset. We trained the model on one K80 GPU for 25,000 iterations with following parameters:
23
+ - learning rate: 1e-3
24
+ - train batch size: 8
25
+ - max input sequence length: 512
26
+ - max target sequence length: 64
27
+
28
+
29
+ ## Model Performance:
30
+
31
+ Model evaluation was done on 2000 evaluation paragraphs of the dataset. Mean [f1 ROUGE scores](https://github.com/pltrdy/rouge) were calculated for the model.
32
+
33
+ | Rouge-1 | Rouge-2 | Rouge-L |
34
+ |---|---|---|
35
+ |0.162 | 0.052 | 0.161 |