EuroEval 15.6.0__py3-none-any.whl → 15.7.0__py3-none-any.whl

euroeval/data_models.py

@@ -388,8 +388,10 @@ class DatasetConfig:
             language.
         _prompt_label_mapping (optional):
             A mapping from the labels to another phrase which is used as a substitute
-            for the label in few-shot evaluation. Defaults to the template for the task
-            and language.
+            for the label in few-shot evaluation. If "auto" then the mapping will be set
+            to a 1:1 mapping between the labels and themselves. If None then the mapping
+            will be set to the default mapping for the task and language. Defaults to
+            None.
         unofficial (optional):
             Whether the dataset is unofficial. Defaults to False.
     """
@@ -405,7 +407,7 @@ class DatasetConfig:
     _num_few_shot_examples: int | None = None
     _max_generated_tokens: int | None = None
     _labels: list[str] | None = None
-    _prompt_label_mapping: dict[str, str] | None = None
+    _prompt_label_mapping: dict[str, str] | t.Literal["auto"] | None = None
     unofficial: bool = False
 
     @property
@@ -475,7 +477,9 @@ class DatasetConfig:
     @property
     def prompt_label_mapping(self) -> dict[str, str]:
         """Mapping from English labels to localised labels."""
-        if self._prompt_label_mapping is not None:
+        if self._prompt_label_mapping == "auto":
+            return {label: label for label in self.labels}
+        elif self._prompt_label_mapping is not None:
             return self._prompt_label_mapping
 
         main_language = self.languages[0]
@@ -517,14 +521,6 @@ class DatasetConfig:
 
         Returns:
             The natural string representation of the labels in specified language.
-
-        Raises:
-            NotImplementedError:
-                If `and_separator` or `or_separator` are `None`, see `Language`.
-
-        Example:
-            >>> get_labels_str(language=DA)
-            "'a', 'b', 'c' eller 'd'"
         """
         main_language = self.languages[0]
 

euroeval/dataset_configs/faroese.py

@@ -12,6 +12,7 @@ FOSENT_CONFIG = DatasetConfig(
     huggingface_id="EuroEval/fosent",
     task=SENT,
     languages=[FO],
+    _num_few_shot_examples=5,
 )
 
 SCALA_FO_CONFIG = DatasetConfig(

euroeval/dataset_configs/finnish.py

@@ -0,0 +1,60 @@
+"""All Finnish dataset configurations used in EuroEval."""
+
+from ..data_models import DatasetConfig
+from ..languages import FI
+from ..tasks import COMMON_SENSE, LA, NER, RC, SENT, SUMM
+
+### Official datasets ###
+
+SCANDISENT_FI_CONFIG = DatasetConfig(
+    name="scandisent-fi",
+    pretty_name="the truncated version of the Finnish part of the binary sentiment "
+    "classification dataset ScandiSent",
+    huggingface_id="EuroEval/scandisent-fi-mini",
+    task=SENT,
+    languages=[FI],
+    _labels=["negative", "positive"],
+)
+
+TURKU_NER_FI_CONFIG = DatasetConfig(
+    name="turku-ner-fi",
+    pretty_name="the Finnish part of the named entity recognition dataset Turku NER",
+    huggingface_id="EuroEval/turku-ner-fi-mini",
+    task=NER,
+    languages=[FI],
+)
+
+TYDIQA_FI_CONFIG = DatasetConfig(
+    name="tydiqa-fi",
+    pretty_name="the Finnish part of the TydiQA reading comprehension dataset",
+    huggingface_id="EuroEval/tydiqa-fi-mini",
+    task=RC,
+    languages=[FI],
+)
+
+XLSUM_FI_CONFIG = DatasetConfig(
+    name="xlsum-fi",
+    pretty_name="the Finnish summarisation dataset XL-Sum",
+    huggingface_id="EuroEval/xlsum-fi-mini",
+    task=SUMM,
+    languages=[FI],
+)
+
+HELLASWAG_FI_CONFIG = DatasetConfig(
+    name="hellaswag-fi",
+    pretty_name="the truncated version of the Finnish common-sense reasoning dataset "
+    "HellaSwag-fi, translated from the English HellaSwag dataset",
+    huggingface_id="EuroEval/hellaswag-fi-mini",
+    task=COMMON_SENSE,
+    languages=[FI],
+)
+
+SCALA_FI_CONFIG = DatasetConfig(
+    name="scala-fi",
+    pretty_name="the Finnish part of the linguistic acceptability dataset ScaLA",
+    huggingface_id="EuroEval/scala-fi",
+    task=LA,
+    languages=[FI],
+)
+
+### Unofficial datasets ###

euroeval/dataset_configs/norwegian.py

@@ -83,6 +83,7 @@ NOR_COMMON_SENSE_QA_CONFIG = DatasetConfig(
     huggingface_id="EuroEval/nor-common-sense-qa",
     task=COMMON_SENSE,
     languages=[NB, NN, NO],
+    _labels=["a", "b", "c", "d", "e"],
 )
 
 
@@ -105,7 +106,6 @@ NORGLM_MULTI_QA = DatasetConfig(
     huggingface_id="EuroEval/norglm-multi-qa",
     task=RC,
     languages=[NB, NN, NO],
-    _num_few_shot_examples=2,
     unofficial=True,
 )
 

euroeval/prompt_templates/linguistic_acceptability.py

@@ -1,7 +1,7 @@
 """Templates for the Linguistic Acceptability task."""
 
 from ..data_models import PromptConfig
-from ..languages import DA, DE, EN, ES, FO, FR, IS, IT, NB, NL, NN, NO, SV
+from ..languages import DA, DE, EN, ES, FI, FO, FR, IS, IT, NB, NL, NN, NO, SV
 
 LA_TEMPLATES = {
     DA: PromptConfig(
@@ -36,6 +36,14 @@ LA_TEMPLATES = {
         default_instruction_prompt="Texto: {text}\n\nDetermina si el texto es "
         "gramaticalmente correcto o no. Responde con {labels_str}, y nada más.",
     ),
+    FI: PromptConfig(
+        default_prompt_label_mapping=dict(correct="kyllä", incorrect="ei"),
+        default_prompt_prefix="Seuraavat ovat lauseita ja ovatko ne "
+        "kieliopillisesti oikein.",
+        default_prompt_template="Lause: {text}\nKieliopillisesti oikein: {label}",
+        default_instruction_prompt="Lause: {text}\n\nMääritä onko lause "
+        "oikein vai ei. Vastaa {labels_str}, ja ei mitään muuta.",
+    ),
     FO: PromptConfig(
         default_prompt_label_mapping=dict(correct="ja", incorrect="nei"),
         default_prompt_prefix="Hetta eru nakrir setningar og um teir eru mállæruliga "

euroeval/prompt_templates/multiple_choice.py

@@ -1,7 +1,7 @@
 """Templates for all multiple choice tasks."""
 
 from ..data_models import PromptConfig
-from ..languages import DA, DE, EN, ES, FR, IS, IT, NB, NL, NN, NO, SV
+from ..languages import DA, DE, EN, ES, FI, FR, IS, IT, NB, NL, NN, NO, SV
 
 # TODO: Missing Faroese
 MULTIPLE_CHOICE_TEMPLATES = {
@@ -36,6 +36,13 @@ MULTIPLE_CHOICE_TEMPLATES = {
         "usando solo {labels_str}, y nada más.",
         default_prompt_label_mapping="auto",
     ),
+    FI: PromptConfig(
+        default_prompt_prefix="Seuraavat ovat monivalintakysymyksiä (vastauksineen).",
+        default_prompt_template="Kysymys: {text}\nVastaus: {label}",
+        default_instruction_prompt="Kysymys: {text}\n\nVastaa yllä olevaan kysymykseen "
+        "käyttämällä {labels_str}, äläkä mitään muuta.",
+        default_prompt_label_mapping="auto",
+    ),
     FR: PromptConfig(
         default_prompt_prefix="Les questions suivantes sont des questions à choix "
         "multiples (avec réponses).",

euroeval/prompt_templates/named_entity_recognition.py

@@ -1,7 +1,7 @@
 """Templates for the Named Entity Recognition task."""
 
 from ..data_models import PromptConfig
-from ..languages import DA, DE, EN, ES, FO, FR, IS, IT, NB, NL, NN, NO, SV
+from ..languages import DA, DE, EN, ES, FI, FO, FR, IS, IT, NB, NL, NN, NO, SV
 
 NER_TEMPLATES = {
     DA: PromptConfig(
@@ -80,6 +80,25 @@ NER_TEMPLATES = {
         "claves {labels_str}. Los valores deben ser listas de las "
         "entidades nombradas de ese tipo, exactamente como aparecen en la oración.",
     ),
+    FI: PromptConfig(
+        default_prompt_label_mapping={
+            "b-per": "henkilö",
+            "i-per": "henkilö",
+            "b-loc": "paikka",
+            "i-loc": "paikka",
+            "b-org": "organisaatio",
+            "i-org": "organisaatio",
+            "b-misc": "muut",
+            "i-misc": "muut",
+        },
+        default_prompt_prefix="Seuraavassa on lauseita ja JSON-sanakirjoja, jotka "
+        "sisältävät annetussa lauseessa esiintyvät nimetyt entiteetit.",
+        default_prompt_template="Lause: {text}\nNimetyt entiteetit: {label}",
+        default_instruction_prompt="Lause: {text}\n\nTunnista lauseessa olevat "
+        "entiteetit. Tulosta ne JSON-sanakirjana, jonka avaimet ovat {labels_str}. "
+        "Arvojen tulee olla listoja kyseisen tyypin nimetyistä entiteeteistä "
+        "täsmälleen siinä muodossa kuin ne esiintyvät lauseessa.",
+    ),
     FO: PromptConfig(
         default_prompt_label_mapping={
             "b-per": "persónur",

euroeval/prompt_templates/reading_comprehension.py

@@ -1,7 +1,7 @@
 """Templates for the Reading Comprehension task."""
 
 from ..data_models import PromptConfig
-from ..languages import DA, DE, EN, ES, FO, FR, IS, IT, NB, NL, NN, NO, SV
+from ..languages import DA, DE, EN, ES, FI, FO, FR, IS, IT, NB, NL, NN, NO, SV
 
 RC_TEMPLATES = {
     DA: PromptConfig(
@@ -39,6 +39,16 @@ RC_TEMPLATES = {
         "sobre el texto anterior en máximo 3 palabras.\n\nPregunta: {question}",
         default_prompt_label_mapping=dict(),
     ),
+    FI: PromptConfig(
+        default_prompt_prefix="Seuraavassa on tekstejä ja niihin liittyviä kysymyksiä "
+        "ja vastauksia.",
+        default_prompt_template="Teksti: {text}\nKysymys: {question} "
+        "\nVastaa enintään 3 sanalla: {label}",
+        default_instruction_prompt="Teksti: {text}\n\nVastaa seuraavaan "
+        "kysymykseen yllä olevasta tekstistä enintään 3 sanalla.\n\n"
+        "Kysymys: {question}",
+        default_prompt_label_mapping=dict(),
+    ),
     FO: PromptConfig(
         default_prompt_prefix="Hetta eru tekstir saman við spurningum og svar.",
         default_prompt_template="Tekstur: {text}\nSpurningur: {question}\nSvara við í "

euroeval/prompt_templates/sentiment_classification.py

@@ -1,7 +1,7 @@
 """Templates for the Sentiment Analysis task."""
 
 from ..data_models import PromptConfig
-from ..languages import DA, DE, EN, ES, FO, FR, IS, IT, NB, NL, NN, NO, SV
+from ..languages import DA, DE, EN, ES, FI, FO, FR, IS, IT, NB, NL, NN, NO, SV
 
 SENT_TEMPLATES = {
     DA: PromptConfig(
@@ -44,6 +44,16 @@ SENT_TEMPLATES = {
         default_instruction_prompt="Documento: {text}\n\nClasifica el sentimiento del "
         "documento. Responde con {labels_str}, y nada más.",
     ),
+    FI: PromptConfig(
+        default_prompt_label_mapping=dict(
+            positive="positiivinen", neutral="neutrali", negative="negatiivinen"
+        ),
+        default_prompt_prefix="Seuraavassa on arvosteluja ja niiden tunnesävy, joka "
+        "voi olla {labels_str}.",
+        default_prompt_template="Teksti: {text}\nTunnesävy: {label}",
+        default_instruction_prompt="Teksti: {text}\n\nLuokittele arvostelun tunnesävy. "
+        "Vastaa vain {labels_str}, ei muuta.",
+    ),
     FO: PromptConfig(
         default_prompt_label_mapping=dict(
             positive="positivt", neutral="neutralt", negative="negativt"

euroeval/prompt_templates/summarization.py

@@ -1,7 +1,7 @@
 """Templates for the Summarization task."""
 
 from ..data_models import PromptConfig
-from ..languages import DA, DE, EN, ES, FR, IS, IT, NB, NL, NN, NO, SV
+from ..languages import DA, DE, EN, ES, FI, FR, IS, IT, NB, NL, NN, NO, SV
 
 # TODO: Missing Faroese
 SUMM_TEMPLATES = {
@@ -36,6 +36,14 @@ SUMM_TEMPLATES = {
         "documento anterior.",
         default_prompt_label_mapping=dict(),
     ),
+    FI: PromptConfig(
+        default_prompt_prefix="Seuraavassa on artikkeleita ja niihin liittyviä "
+        "tiivistelmiä.",
+        default_prompt_template="Uutisartikkeli: {text}\nTiivistelmä: {target_text}",
+        default_instruction_prompt="Uutisartikkeli: {text}\n\nKirjoita tiivistelmä "
+        "yllä olevasta artikkelista.",
+        default_prompt_label_mapping=dict(),
+    ),
     FR: PromptConfig(
         default_prompt_prefix="Les documents suivants sont accompagnés d'un résumé.",
         default_prompt_template="Document: {text}\nRésumé: {target_text}",

euroeval/task_group_utils/sequence_classification.py

@@ -132,22 +132,23 @@ def extract_labels_from_generation(
         The predicted labels.
     """
     if model_output.scores is not None:
-        return get_closest_logprobs_labels(
+        labels = get_closest_logprobs_labels(
             generation_logprobs=model_output.scores,
             dataset_config=dataset_config,
             first_label_token_mapping=first_label_token_mapping,
         )
-    else:
-        return get_closest_word_edit_labels(
-            generated_sequences=model_output.sequences, dataset_config=dataset_config
-        )
+        if labels is not None:
+            return labels
+    return get_closest_word_edit_labels(
+        generated_sequences=model_output.sequences, dataset_config=dataset_config
+    )
 
 
 def get_closest_logprobs_labels(
     generation_logprobs: list[list[list[tuple[str, float]]]],
     dataset_config: "DatasetConfig",
     first_label_token_mapping: dict[str, str] | bool,
-) -> list[str]:
+) -> list[str] | None:
     """Get the labels with the highest predicted logprob value.
 
     In case a candidate label is split into multiple tokens, we only use the first
@@ -167,7 +168,7 @@ def get_closest_logprobs_labels(
             mapping is outputted then the model will always output scores).
 
     Returns:
-        The predicted labels.
+        The predicted labels, or None if labels could not be extracted.
 
     Raises:
         InvalidBenchmark:
@@ -193,10 +194,7 @@ def get_closest_logprobs_labels(
             # We want to use the first generated label which contains a unique candidate
             # label, as the output label
             output_label: str | None = None
-            previously_generated_labels: list[str] = list()
-            for label_idx, generated_label in enumerate(generated_labels):
-                generated_label = "".join(previously_generated_labels) + generated_label
-
+            for generated_label in generated_labels:
                 # Get the candidate labels that starts with the generated label
                 if isinstance(first_label_token_mapping, dict):
                     if any(
@@ -222,31 +220,28 @@ def get_closest_logprobs_labels(
                         if candidate_label.startswith(generated_label)
                     }
 
-                # If we can uniquely determine the output label, we break the loop. If
-                # there are multiple possible labels then we store the current one, and
-                # concatenate it with the next generated label. We can only do this if
-                # the current one is the first one, however, since we're using greedy
-                # sampling. In case this happens for a label that is not the first one,
-                # we warn the user.
+                # If we can uniquely determine the output label, we break the loop.
                 if len(candidate_output_labels) == 1:
                     output_label = candidate_output_labels.pop()
                     break
+
+                # If we have multiple candidate labels, we cannot uniquely determine the
+                # output label, so we abandon extracting the labels using logprobs and
+                # fall back to using word edit distance.
                 elif len(candidate_output_labels) > 1:
-                    if label_idx == 0:
-                        previously_generated_labels.append(generated_label)
-                    else:
-                        output_label = candidate_output_labels.pop()
-                        candidate_output_labels.add(output_label)
-                        raise InvalidBenchmark(
-                            "Multiple candidate labels found for the generated label "
-                            f"{generated_label!r}: {candidate_output_labels}. Since "
-                            "this is not the first generated label, we cannot "
-                            "concatenate it with the next generated label. We are thus "
-                            f"forced to use the arbitrary {output_label!r} as the "
-                            "output label, potentially resulting in worse performance. "
-                            "Please report this issue to the EuroEval team at "
-                            "github.com/EuroEval/EuroEval/issues."
-                        )
+                    log_once(
+                        "Multiple candidate labels found for the generated label "
+                        f"{generated_label!r}: {candidate_output_labels}. This means "
+                        "that using logprobs to extract the labels is not reliable, "
+                        "and we will instead fall back to extracting the labels "
+                        "using word edit distance.",
+                        level=logging.DEBUG,
+                    )
+                    return None
+
+                # If no candidate label is found, we ignore the generated label, as it
+                # basically means that the model is just really bad at generating
+                # labels.
                 elif len(candidate_output_labels) == 0:
                     logger.debug(
                         f"No candidate label found for the generated label "

euroeval/task_group_utils/text_to_text.py

@@ -10,11 +10,7 @@ from evaluate import EvaluationModule
 from ..constants import METRIC_ATTRIBUTES_TAKING_UP_MEMORY
 from ..data_models import BenchmarkConfig, DatasetConfig, GenerativeModelOutput
 from ..exceptions import InvalidBenchmark
-from ..utils import (
-    HiddenPrints,
-    clear_memory,
-    raise_if_model_output_contains_nan_values,
-)
+from ..utils import HiddenPrints, raise_if_model_output_contains_nan_values
 
 if t.TYPE_CHECKING:
     from transformers.trainer_utils import EvalPrediction
@@ -89,20 +85,8 @@ def compute_metrics(
                     score_dict: dict[str, float] | None = metric.compute(
                         predictions=predictions, references=labels, **cfg.compute_kwargs
                     )
-
-                # Clear the cache of the BERTScorer to avoid memory leaks
-                for attribute in METRIC_ATTRIBUTES_TAKING_UP_MEMORY:
-                    if hasattr(metric, attribute):
-                        delattr(metric, attribute)
-
-                clear_memory()
                 break
             except Exception as e:
-                # Clear the cache of the BERTScorer to avoid memory leaks
-                if hasattr(metric, "cached_bertscorer"):
-                    del metric.cached_bertscorer
-                    clear_memory()
-
                 oom_error = [
                     "CUDA out of memory",
                     "CUDA error",
@@ -111,16 +95,7 @@ def compute_metrics(
                 if not any(error in str(e) for error in oom_error):
                     raise InvalidBenchmark(str(e))
 
-                if cfg.compute_kwargs.get("batch_size", 1) > 1:
-                    batch_size = cfg.compute_kwargs["batch_size"]
-                    cfg.compute_kwargs["batch_size"] = batch_size // 2
-                    logger.debug(
-                        "Out of memory error occurred during the computation of "
-                        f"the metric {cfg.pretty_name}. Reducing the batch size to "
-                        f"{cfg.compute_kwargs['batch_size']}."
-                    )
-                elif cfg.compute_kwargs.get("device", "cpu") != "cpu":
-                    cfg.compute_kwargs["batch_size"] = 32
+                if cfg.compute_kwargs.get("device", "cpu") != "cpu":
                     cfg.compute_kwargs["device"] = "cpu"
                     logger.debug(
                         "Out of memory error occurred during the computation of "
@@ -129,6 +104,14 @@ def compute_metrics(
                     )
                 else:
                     raise InvalidBenchmark(str(e))
+            finally:
+                for attribute in METRIC_ATTRIBUTES_TAKING_UP_MEMORY:
+                    if hasattr(metric, attribute):
+                        logger.debug(
+                            f"Deleting the {attribute!r} attribute of the metric "
+                            f"{cfg.pretty_name} to free up memory."
+                        )
+                        delattr(metric, attribute)
 
         # The metric returns None if we are running on multi-GPU and the current
         # process is not the main process

euroeval/tasks.py

@@ -142,7 +142,7 @@ SUMM = Task(
             huggingface_id="bertscore",
             results_key="f1",
             compute_kwargs=dict(
-                model_type="microsoft/mdeberta-v3-base", device="auto", batch_size=32
+                model_type="microsoft/mdeberta-v3-base", device="auto", batch_size=1
             ),
         ),
         MetricConfig(

euroeval/tokenization_utils.py

@@ -7,6 +7,7 @@ import typing as t
 import torch
 
 from .constants import TASK_GROUPS_USING_LOGPROBS
+from .enums import GenerativeType
 from .exceptions import InvalidModel
 from .utils import log_once
 
@@ -14,7 +15,7 @@ if t.TYPE_CHECKING:
     from transformers.tokenization_utils import PreTrainedTokenizer
     from transformers.tokenization_utils_base import PreTrainedTokenizerBase
 
-    from .data_models import DatasetConfig
+    from .data_models import DatasetConfig, ModelConfig
 
 
 logger = logging.getLogger("euroeval")
@@ -254,35 +255,50 @@ def get_end_of_chat_token_ids(tokenizer: "PreTrainedTokenizer") -> list[int] | N
 
 
 def get_first_label_token_mapping(
-    dataset_config: "DatasetConfig", tokenizer: "PreTrainedTokenizer | None"
+    dataset_config: "DatasetConfig",
+    model_config: "ModelConfig",
+    tokenizer: "PreTrainedTokenizer | None",
+    generative_type: "GenerativeType | None",
 ) -> dict[str, str] | bool:
     """Check if the model should output scores.
 
     Args:
         dataset_config:
             The dataset configuration.
+        model_config:
+            The model configuration.
         tokenizer:
             The tokenizer, or None if not available.
+        generative_type:
+            The generative type, or None if not available.
 
     Returns:
         A mapping from labels to the first token in each label, or alternatively a
         Boolean value indicating whether the model should output scores (if the mapping
         is outputted then the model will always output scores).
     """
+    if generative_type == GenerativeType.REASONING:
+        log_once(
+            f"The model {model_config.model_id!r} is a reasoning model and "
+            "thus does not support logprobs, so we do not enable it.",
+            level=logging.DEBUG,
+        )
+        return False
+
     # If we do not have any tokenizer, then we cannot check if the model should output
     # scores and we just assume it should if the dataset supports it
     output_scores = dataset_config.task.task_group in TASK_GROUPS_USING_LOGPROBS
     if tokenizer is None:
         if output_scores:
             log_once(
-                "The model will output scores, since the dataset supports it and no "
-                "tokenizer is available.",
+                f"The model {model_config.model_id!r} will output scores, since the "
+                "dataset supports it and no tokenizer is available.",
                 level=logging.DEBUG,
             )
         else:
             log_once(
-                "The model will not output scores, since the dataset does not support "
-                "it and no tokenizer is available.",
+                f"The model {model_config.model_id!r} will not output scores, since "
+                "the dataset does not support it and no tokenizer is available.",
                 level=logging.DEBUG,
             )
         return output_scores

{euroeval-15.6.0.dist-info → euroeval-15.7.0.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: EuroEval
-Version: 15.6.0
+Version: 15.7.0
 Summary: The robust European language model benchmark.
 Project-URL: Repository, https://github.com/EuroEval/EuroEval
 Project-URL: Issues, https://github.com/EuroEval/EuroEval/issues
@@ -32,7 +32,7 @@ Requires-Python: <4.0,>=3.10
 Requires-Dist: accelerate>=0.34.2
 Requires-Dist: bert-score>=0.3.13
 Requires-Dist: click>=8.1.3
-Requires-Dist: datasets>=2.15.0
+Requires-Dist: datasets>=3.5.0
 Requires-Dist: demjson3>=3.0.6
 Requires-Dist: evaluate>=0.4.1
 Requires-Dist: huggingface-hub>=0.30.1
@@ -237,6 +237,19 @@ A huge thank you to all the contributors who have helped make this project a suc
 <a href="https://github.com/ThomasKluiters"><img src="https://avatars.githubusercontent.com/u/8137941" width=50 alt="Contributor avatar for ThomasKluiters"/></a>
 <a href="https://github.com/BramVanroy"><img src="https://avatars.githubusercontent.com/u/2779410" width=50 alt="Contributor avatar for BramVanroy"/></a>
 <a href="https://github.com/peregilk"><img src="https://avatars.githubusercontent.com/u/9079808" width=50 alt="Contributor avatar for peregilk"/></a>
+<a href="https://github.com/Rijgersberg"><img src="https://avatars.githubusercontent.com/u/8604946" width=50 alt="Contributor avatar for Rijgersberg"/></a>
+
+
+### Contribute to EuroEval
+
+We welcome contributions to EuroEval! Whether you're fixing bugs, adding features, or
+contributing new datasets, your help makes this project better for everyone.
+
+- **General contributions**: Check out our [contribution guidelines](CONTRIBUTING.md)
+  for information on how to get started.
+- **Adding datasets**: If you're interested in adding a new dataset to EuroEval, we have
+  a [dedicated guide](NEW_DATASET_GUIDE.md) with step-by-step instructions.
+
 
 ### Special Thanks
 - Thanks to [Google](https://google.com/) for sponsoring Gemini credits as part of their