Evaluation of pretrained language models on mono- or multilingual language tasks.

---

Maintainers

• Dan Saattrup Nielsen (@saattrupdan, [email protected])
• Kenneth Enevoldsen (@KennethEnevoldsen, [email protected])

Installation

To install the package simply write the following command in your favorite terminal:

$ pip install scandeval[all]

This will install the ScandEval package with all extras. You can also install the minimal version by leaving out the [all], in which case the package will let you know when an evaluation requires a certain extra dependency, and how you install it.

Quickstart

Benchmarking from the Command Line

The easiest way to benchmark pretrained models is via the command line interface. After having installed the package, you can benchmark your favorite model like so:

$ scandeval --model <model-id>

Here model is the HuggingFace model ID, which can be found on the HuggingFace Hub. By default this will benchmark the model on all the tasks available. If you want to benchmark on a particular task, then use the --task argument:

$ scandeval --model <model-id> --task sentiment-classification

We can also narrow down which languages we would like to benchmark on. This can be done by setting the --language argument. Here we thus benchmark the model on the Danish sentiment classification task:

$ scandeval --model <model-id> --task sentiment-classification --language da

Multiple models, datasets and/or languages can be specified by just attaching multiple arguments. Here is an example with two models:

$ scandeval --model <model-id1> --model <model-id2>

The specific model version/revision to use can also be added after the suffix '@':

$ scandeval --model <model-id>@<commit>

This can be a branch name, a tag name, or a commit id. It defaults to 'main' for latest.

See all the arguments and options available for the scandeval command by typing

$ scandeval --help

Benchmarking from a Script

In a script, the syntax is similar to the command line interface. You simply initialise an object of the Benchmarker class, and call this benchmark object with your favorite model:

>>> from scandeval import Benchmarker
>>> benchmark = Benchmarker()
>>> benchmark(model="<model>")

To benchmark on a specific task and/or language, you simply specify the task or language arguments, shown here with same example as above:

>>> benchmark(model="<model>", task="sentiment-classification", language="da")

If you want to benchmark a subset of all the models on the Hugging Face Hub, you can simply leave out the model argument. In this example, we're benchmarking all Danish models on the Danish sentiment classification task:

>>> benchmark(task="sentiment-classification", language="da")

Benchmarking from Docker

A Dockerfile is provided in the repo, which can be downloaded and run, without needing to clone the repo and installing from source. This can be fetched programmatically by running the following:

$ wget https://raw.githubusercontent.com/ScandEval/ScandEval/main/Dockerfile.cuda

Next, to be able to build the Docker image, first ensure that the NVIDIA Container Toolkit is installed and configured. Ensure that the the CUDA version stated at the top of the Dockerfile matches the CUDA version installed (which you can check using nvidia-smi). After that, we build the image as follows:

$ docker build --pull -t scandeval -f Dockerfile.cuda .

With the Docker image built, we can now evaluate any model as follows:

$ docker run -e args="<scandeval-arguments>" --gpus 1 --name scandeval --rm scandeval

Here <scandeval-arguments> consists of the arguments added to the scandeval CLI argument. This could for instance be --model <model-id> --task sentiment-classification.

Special Thanks 🙏

• Thanks to OpenAI for sponsoring OpenAI credits as part of their Researcher Access Program.
• Thanks to UWV and KU Leuven for sponsoring the Azure OpenAI credits used to evaluate GPT-4-turbo in Dutch.
• Thanks to Miðeind for sponsoring the OpenAI credits used to evaluate GPT-4-turbo in Icelandic and Faroese.
• Thanks to CHC for sponsoring the OpenAI credits used to evaluate GPT-4-turbo in German.

Citing ScandEval

If you want to cite the framework then feel free to use this:

@article{nielsen2024encoder,
  title={Encoder vs Decoder: Comparative Analysis of Encoder and Decoder Language Models on Multilingual NLU Tasks},
  author={Nielsen, Dan Saattrup and Enevoldsen, Kenneth and Schneider-Kamp, Peter},
  journal={arXiv preprint arXiv:2406.13469},
  year={2024}
}
@inproceedings{nielsen2023scandeval,
  author = {Nielsen, Dan Saattrup},
  booktitle = {Proceedings of the 24th Nordic Conference on Computational Linguistics (NoDaLiDa)},
  month = may,
  pages = {185--201},
  title = {{ScandEval: A Benchmark for Scandinavian Natural Language Processing}},
  year = {2023}
}

Remarks

The image used in the logo has been created by the amazing Scandinavia and the World team. Go check them out!