manuelciosici
commited on
Commit
•
c349228
1
Parent(s):
3273749
Add README
Browse files
README.md
ADDED
@@ -0,0 +1,55 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# dant5-large
|
2 |
+
|
3 |
+
---
|
4 |
+
language:
|
5 |
+
- da
|
6 |
+
tags:
|
7 |
+
- t5
|
8 |
+
license: cc-by-4.0
|
9 |
+
datasets:
|
10 |
+
- dagw
|
11 |
+
widget:
|
12 |
+
- text: "Aarhus er Danmarks <extra_id_0>.<extra_id_2>"
|
13 |
+
co2_eq_emissions:
|
14 |
+
training_type: "pretraining"
|
15 |
+
geographical_location: "Copenhagen, Denmark"
|
16 |
+
hardware_used: "4 A100 GPUs, 508 training hours"
|
17 |
+
---
|
18 |
+
|
19 |
+
`dant5-large` is a 770M parameter model with architecture identical to `t5-large`. It was trained for 10 epochs on the Danigh GigaWord Corpus ([official website](https://gigaword.dk), [paper](https://aclanthology.org/2021.nodalida-main.46/)).
|
20 |
+
|
21 |
+
## To use the model
|
22 |
+
|
23 |
+
```python
|
24 |
+
from transformers import AutoTokenizer, T5ForConditionalGeneration
|
25 |
+
|
26 |
+
model_name = "strombergnlp/dant5-large"
|
27 |
+
|
28 |
+
tokenizer = AutoTokenizer.from_pretrained(model_name)
|
29 |
+
model = T5ForConditionalGeneration.from_pretrained(model_name)
|
30 |
+
|
31 |
+
original_text = "Aarhus er Danmarks <extra_id_0> landets ældste. Under navnet Aros, som betyder å-munding, optræder den i skriftlige kilder i 900-tallet, men <extra_id_1> historie tilbage til 700-tallet.<extra_id_2>"
|
32 |
+
original_label = "<extra_id_0> næststørste by og en af <extra_id_1> arkæologiske fund fører dens <extra_id_2>"
|
33 |
+
input_ids = tokenizer(original_text, return_tensors="pt").input_ids
|
34 |
+
labels = tokenizer(original_label, return_tensors="pt").input_ids
|
35 |
+
|
36 |
+
loss = model(input_ids=input_ids, labels=labels).loss
|
37 |
+
print(f"Original text: {original_text}")
|
38 |
+
print(f"Original label: {original_label}")
|
39 |
+
print(f"Loss for the original label is {loss.item()}")
|
40 |
+
|
41 |
+
sequence_ids = model.generate(input_ids)
|
42 |
+
sequences = tokenizer.batch_decode(sequence_ids)
|
43 |
+
print(f"A sample generated continuation: ")
|
44 |
+
print(sequences[0])
|
45 |
+
```
|
46 |
+
|
47 |
+
You should see output similar to:
|
48 |
+
|
49 |
+
```
|
50 |
+
Original text: Aarhus er Danmarks <extra_id_0> landets ældste. Under navnet Aros, som betyder å-munding, optræder den i skriftlige kilder i 900-tallet, men <extra_id_1> historie tilbage til 700-tallet.<extra_id_2>
|
51 |
+
Original label: <extra_id_0> næststørste by og en af <extra_id_1> arkæologiske fund fører dens <extra_id_2>
|
52 |
+
Loss for the original label is 4.174272537231445
|
53 |
+
A sample generated continuation:
|
54 |
+
<pad><extra_id_0> ældste by og<extra_id_1> har sin<extra_id_2> Se også<extra_id_3></s>
|
55 |
+
```
|