Hugging Face's logo

flax-community
/
dansk-gpt-wiki

Text Generation
Transformers
PyTorch
JAX
TensorBoard
Safetensors
Danish
gpt2
text-generation-inference
Inference Endpoints
Model card Files Metrics Community
3
dansk-gpt-wiki
File size: 2,903 Bytes 
1d2d61e
 
 
 
 
 
 
 
 
 
 
 
b452553
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1d2d61e
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
b452553
 
 
 
 
 
 
 
 
 
 
 
1d2d61e
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
 
---
language: da
widget:
- text: "Jeg elsker livet"
---

# GPT2-svenska-wikipedia
A Danish GPT2 style model trained using Flax CLM pipeline on the Danish
part of the wiki40b dataset.

https://huggingface.co/datasets/wiki40b

## Model series
This model is part of a series of models training on TPU with Flax Jax during Huggingface Flax/Jax challenge.

## Gpt models

## Swedish Gpt
https://huggingface.co/birgermoell/swedish-gpt/

## Swedish gpt wiki
https://huggingface.co/flax-community/swe-gpt-wiki

# Nordic gpt wiki
https://huggingface.co/flax-community/nordic-gpt-wiki

## Dansk gpt wiki
https://huggingface.co/flax-community/dansk-gpt-wiki

## Norsk gpt wiki
https://huggingface.co/flax-community/norsk-gpt-wiki

## Roberta models

## Nordic Roberta Wiki
https://huggingface.co/flax-community/nordic-roberta-wiki

## Swe Roberta Wiki Oscar
https://huggingface.co/flax-community/swe-roberta-wiki-oscar

## Roberta Swedish Scandi
https://huggingface.co/birgermoell/roberta-swedish-scandi

## Roberta Swedish
https://huggingface.co/birgermoell/roberta-swedish

## Swedish T5 model
https://huggingface.co/birgermoell/t5-base-swedish




## Data cleaning and preprocessing
The data was cleaned and preprocessed using the following script. Make sure to install depencies for beam_runner to make the dataset work.

```python
from datasets import load_dataset
def load_and_clean_wiki():
    dataset = load_dataset('wiki40b', 'da', beam_runner='DirectRunner', split="train")
    #dataset = load_dataset('wiki40b', 'sv', beam_runner='DirectRunner')
    dataset = dataset.remove_columns(['wikidata_id', 'version_id'])
    filtered_dataset = dataset.map(filter_wikipedia)
    # filtered_dataset[:3]
    # print(filtered_dataset[:3])
    return filtered_dataset

def filter_wikipedia(batch):
    batch["text"] = " ".join(batch["text"].split("\
_START_SECTION_\
"))
    batch["text"] = " ".join(batch["text"].split("\
_START_ARTICLE_\
"))
    batch["text"] = " ".join(batch["text"].split("\
_START_ARTICLE_\
"))
    batch["text"] = " ".join(batch["text"].split("\
_START_PARAGRAPH_\
"))
    batch["text"] = " ".join(batch["text"].split("_NEWLINE_"))
    batch["text"] = " ".join(batch["text"].split("\xa0"))
    return batch
```

## Training script
The following training script was used to train the model.
```bash
./run_clm_flax.py     --output_dir="${MODEL_DIR}"     --model_type="gpt2"     --config_name="${MODEL_DIR}"     --tokenizer_name="${MODEL_DIR}"     --dataset_name="wiki40b"     --dataset_config_name="da"     --do_train --do_eval     --block_size="512"     --per_device_train_batch_size="64"     --per_device_eval_batch_size="64"     --learning_rate="5e-3" --warmup_steps="1000"     --adam_beta1="0.9" --adam_beta2="0.98" --weight_decay="0.01"     --overwrite_output_dir     --num_train_epochs="20"     --logging_steps="500"     --save_steps="1000"     --eval_steps="2500"     --push_to_hub
```