README.md

---
language:
- 'no'
- nb
- nn
inference: false
tags:
- T5
- NorT5
- Norwegian
- encoder-decoder
license: apache-2.0
---

# NorT5 large

<img src="https://huggingface.co/ltg/norbert3-base/resolve/main/norbert.png" width=12.5%>

The official release of a new generation of NorT5 language models described in paper [**NorBench — A Benchmark for Norwegian Language Models**](https://arxiv.org/abs/2305.03880). Plese read the paper to learn more details about the model.


## Other sizes:
- [NorT5 xs (32M)](https://huggingface.co/ltg/nort5-xs)
- [NorT5 small (88M)](https://huggingface.co/ltg/nort5-small)
- [NorT5 base (228M)](https://huggingface.co/ltg/nort5-base)
- [NorT5 large (808M)](https://huggingface.co/ltg/nort5-large)


## Encoder-only NorBERT siblings:
- [NorBERT 3 xs (15M)](https://huggingface.co/ltg/norbert3-xs)
- [NorBERT 3 small (40M)](https://huggingface.co/ltg/norbert3-small)
- [NorBERT 3 base (123M)](https://huggingface.co/ltg/norbert3-base)
- [NorBERT 3 large (323M)](https://huggingface.co/ltg/norbert3-large)


## Example usage

This model currently needs a custom wrapper from `modeling_nort5.py`, you should therefore load the model with `trust_remote_code=True`.


```python
import torch
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("ltg/nort5-large")
model = AutoModelForSeq2SeqLM.from_pretrained("ltg/nort5-large", trust_remote_code=True)


# MASKED LANGUAGE MODELING

sentence = "Brukseksempel: Elektrisk oppvarming. Definisjonen på ordet oppvarming er[MASK_0]."
encoding = tokenizer(sentence)

input_tensor = torch.tensor([encoding.input_ids])
output_tensor = model.generate(input_tensor, decoder_start_token_id=7, eos_token_id=8)
tokenizer.decode(output_tensor.squeeze(), skip_special_tokens=True)

# should output:  å varme opp


# PREFIX LANGUAGE MODELING
# you need to finetune this model or use `nort5-{size}-lm` model, which is finetuned on prefix language modeling

sentence = "Brukseksempel: Elektrisk oppvarming. Definisjonen på ordet oppvarming er (Wikipedia) "
encoding = tokenizer(sentence)

input_tensor = torch.tensor([encoding.input_ids])
output_tensor = model.generate(input_tensor, max_new_tokens=50, num_beams=4, do_sample=False)
tokenizer.decode(output_tensor.squeeze())

# should output: [BOS]ˈoppvarming, det vil si at det skjer en endring i temperaturen i et medium, f.eks. en ovn eller en radiator, slik at den blir varmere eller kaldere, eller at den blir varmere eller kaldere, eller at den blir
```


The following classes are currently implemented: `AutoModel`, `AutoModelForSeq2SeqLM`.

## Cite us

```bibtex
@inproceedings{samuel-etal-2023-norbench,
    title = "{N}or{B}ench {--} A Benchmark for {N}orwegian Language Models",
    author = "Samuel, David  and
      Kutuzov, Andrey  and
      Touileb, Samia  and
      Velldal, Erik  and
      {\O}vrelid, Lilja  and
      R{\o}nningstad, Egil  and
      Sigdel, Elina  and
      Palatkina, Anna",
    booktitle = "Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa)",
    month = may,
    year = "2023",
    address = "T{\'o}rshavn, Faroe Islands",
    publisher = "University of Tartu Library",
    url = "https://aclanthology.org/2023.nodalida-1.61",
    pages = "618--633",
    abstract = "We present NorBench: a streamlined suite of NLP tasks and probes for evaluating Norwegian language models (LMs) on standardized data splits and evaluation metrics. We also introduce a range of new Norwegian language models (both encoder and encoder-decoder based). Finally, we compare and analyze their performance, along with other existing LMs, across the different benchmark tests of NorBench.",
}

```