initial commit

README.md

@@ -0,0 +1,97 @@
+---
+language: "da"
+tags:
+- ælæctra
+- pytorch
+- danish
+- ELECTRA-Small
+- replaced token detection
+license: "mit"
+datasets:
+- DAGW
+metrics:
+- f1
+---
+
+# Ælæctra - A Step Towards More Efficient Danish Natural Language Processing
+**Ælæctra** is a Danish Transformer-based language model created to enhance the variety of Danish NLP resources with a more efficient model compared to previous state-of-the-art (SOTA) models. Initially a cased and an uncased model are released. It was created as part of a Cognitive Science bachelor's thesis.
+
+Ælæctra was pretrained with the ELECTRA-Small (Clark et al., 2020) pretraining approach by using the Danish Gigaword Corpus (Strømberg-Derczynski et al., 2020) and evaluated on Named Entity Recognition (NER) tasks. Since NER only presents a limited picture of Ælæctra's capabilities I am very interested in further evaluations. Therefore, if you employ it for any task, feel free to hit me up your findings!
+
+Ælæctra was, as mentioned, created to enhance the Danish NLP capabilties and please do note how this GitHub still does not support the Danish characters "*Æ, Ø and Å*" as the title of this repository becomes "*-l-ctra*". How ironic.🙂
+
+Here is an example on how to load both the cased and the uncased Ælæctra model in [PyTorch](https://pytorch.org/) using the [🤗Transformers](https://github.com/huggingface/transformers) library:
+
+```python
+from transformers import AutoTokenizer, AutoModelForPreTraining
+
+tokenizer = AutoTokenizer.from_pretrained("Maltehb/-l-ctra-cased")
+model = AutoModelForPreTraining.from_pretrained("Maltehb/-l-ctra-cased")
+```
+
+```python
+from transformers import AutoTokenizer, AutoModelForPreTraining
+
+tokenizer = AutoTokenizer.from_pretrained("Maltehb/-l-ctra-uncased")
+model = AutoModelForPreTraining.from_pretrained("Maltehb/-l-ctra-uncased")
+```
+
+### Evaluation of current Danish Language Models 
+
+Ælæctra, Danish BERT (DaBERT) and multilingual BERT (mBERT) were evaluated:
+
+| Model | Layers | Hidden Size | Params | AVG NER micro-f1 (DaNE-testset) | Average Inference Time (Sec/Epoch) | Download | 
+| --- | --- | --- | --- | ---  | --- | --- |
+| Ælæctra Uncased | 12 | 256 | 13.7M | 78.03 (SD = 1.28) | 10.91 | [Link for model](https://www.dropbox.com/s/cag7prs1nvdchqs/%C3%86l%C3%A6ctra.zip?dl=0) | 
+| Ælæctra Cased | 12 | 256 | 14.7M | 80.08 (SD = 0.26) | 10.92 | [Link for model](https://www.dropbox.com/s/cag7prs1nvdchqs/%C3%86l%C3%A6ctra.zip?dl=0) | 
+| DaBERT | 12 | 768 | 110M | 84.89 (SD = 0.64) | 43.03 | [Link for model](https://www.dropbox.com/s/19cjaoqvv2jicq9/danish_bert_uncased_v2.zip?dl=1) | 
+| mBERT Uncased | 12 | 768 | 167M | 80.44 (SD = 0.82) | 72.10 | [Link for model](https://storage.googleapis.com/bert_models/2018_11_03/multilingual_L-12_H-768_A-12.zip) | 
+| mBERT Cased | 12 | 768 | 177M | 83.79 (SD = 0.91) | 70.56 | [Link for model](https://storage.googleapis.com/bert_models/2018_11_23/multi_cased_L-12_H-768_A-12.zip) | 
+
+
+On [DaNE](https://danlp.alexandra.dk/304bd159d5de/datasets/ddt.zip) (Hvingelby et al., 2020), Ælæctra scores slightly worse than both cased and uncased Multilingual BERT (Devlin et al., 2019) and Danish BERT (Danish BERT, 2019/2020), however, is more than 3 times faster per batch at inference time. For a full description of the evaluation and specification of the model read the thesis: 'Ælæctra - A Step Towards More Efficient Danish Natural Language Processing'. 
+
+### Pretraining
+To pretrain Ælæctra it is recommended to build a Docker Container from the [Dockerfile](https://github.com/MalteHB/Ælæctra/tree/master/notebooks/fine-tuning/). Next, simply follow the [pretraining notebooks](https://github.com/MalteHB/Ælæctra/tree/master/infrastructure/Dockerfile/) 
+
+The pretraining was done by utilizing a single NVIDIA Tesla V100 GPU with 16 GiB, endowed by the Danish data company [KMD](https://www.kmd.dk/). The pretraining took approximately 4 days and 9.5 hours for both the cased and uncased model
+
+### Fine-tuning
+To fine-tune any Ælæctra model follow the [fine-tuning notebooks](https://github.com/MalteHB/Ælæctra/tree/master/notebooks/fine-tuning/)
+
+### References
+Clark, K., Luong, M.-T., Le, Q. V., & Manning, C. D. (2020). ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators. ArXiv:2003.10555 [Cs]. http://arxiv.org/abs/2003.10555
+
+Danish BERT. (2020). BotXO. https://github.com/botxo/nordic_bert (Original work published 2019)
+
+Devlin, J., Chang, M.-W., Lee, K., & Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. ArXiv:1810.04805 [Cs]. http://arxiv.org/abs/1810.04805
+
+Hvingelby, R., Pauli, A. B., Barrett, M., Rosted, C., Lidegaard, L. M., & Søgaard, A. (2020). DaNE: A Named Entity Resource for Danish. Proceedings of the 12th Language Resources and Evaluation Conference, 4597–4604. https://www.aclweb.org/anthology/2020.lrec-1.565
+
+Strømberg-Derczynski, L., Baglini, R., Christiansen, M. H., Ciosici, M. R., Dalsgaard, J. A., Fusaroli, R., Henrichsen, P. J., Hvingelby, R., Kirkedal, A., Kjeldsen, A. S., Ladefoged, C., Nielsen, F. Å., Petersen, M. L., Rystrøm, J. H., & Varab, D. (2020). The Danish Gigaword Project. ArXiv:2005.03521 [Cs]. http://arxiv.org/abs/2005.03521
+
+
+#### Acknowledgements
+As the majority of this repository is build upon [the works](https://github.com/google-research/electra) by the team at Google who created ELECTRA, a HUGE thanks to them is in order. 
+
+A Giga thanks also goes out to the incredible people who collected The Danish Gigaword Corpus (Strømberg-Derczynski et al., 2020). 
+
+Furthermore, I would like to thank my supervisor [Riccardo Fusaroli](https://github.com/fusaroli) for the support with the thesis, and a special thanks goes out to [Kenneth Enevoldsen](https://github.com/KennethEnevoldsen) for his continuous feedback. 
+
+Lastly, i would like to thank KMD, my colleagues from KMD, and my peers and co-students from Cognitive Science for encouriging me to keep on working hard and holding my head up high!
+
+#### Contact
+
+For help or further information feel free to connect with the author Malte Højmark-Bertelsen on [hjb@kmd.dk](mailto:hjb@kmd.dk?subject=[GitHub]%20Ælæctra) or any of the following platforms:
+
+[<img align="left" alt="MalteHB | Twitter" width="22px" src="https://cdn.jsdelivr.net/npm/simple-icons@v3/icons/twitter.svg" />][twitter]
+[<img align="left" alt="MalteHB | LinkedIn" width="22px" src="https://cdn.jsdelivr.net/npm/simple-icons@v3/icons/linkedin.svg" />][linkedin]
+[<img align="left" alt="MalteHB | Instagram" width="22px" src="https://cdn.jsdelivr.net/npm/simple-icons@v3/icons/instagram.svg" />][instagram]
+
+<br />
+
+</details>
+
+[twitter]: https://twitter.com/malteH_B
+[instagram]: https://www.instagram.com/maltemusen/
+[linkedin]: https://www.linkedin.com/in/malte-h%C3%B8jmark-bertelsen-9a618017b/

config.json

@@ -0,0 +1,25 @@
+{
+  "architectures": [
+    "ElectraForPreTraining"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "embedding_size": 128,
+  "generator_size": "0.25",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 256,
+  "initializer_range": 0.02,
+  "intermediate_size": 1024,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "electra",
+  "num_attention_heads": 4,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "summary_activation": "gelu",
+  "summary_last_dropout": 0.1,
+  "summary_type": "first",
+  "summary_use_proj": true,
+  "type_vocab_size": 2,
+  "vocab_size": 32000
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e0955ca630cff4aa08e2bb22a36f1fe0cf37a81922ae96f5d6594429bd502180
+size 57979406

special_tokens_map.json

@@ -0,0 +1 @@
+{"unk_token": "[UNK]", "sep_token": "[SEP]", "pad_token": "[PAD]", "cls_token": "[CLS]", "mask_token": "[MASK]"}

tokenizer_config.json

@@ -0,0 +1 @@
+{"do_lower_case": false, "special_tokens_map_file": null, "full_tokenizer_file": null}