improve dataset card: add paper link, badges, task categories, usage example

README.md

@@ -91,10 +91,10 @@ configs:
     path: french_boolq/french_boolq_test.jsonl
 - config_name: rte3-french
   data_files:
-    - split: validation
-      path: rte3-french/*dev*.jsonl
-    - split: test
-      path: rte3-french/*test*.jsonl
+  - split: validation
+    path: rte3-french/*dev*.jsonl
+  - split: test
+    path: rte3-french/*test*.jsonl
 - config_name: wino_x_lm
   data_files:
   - split: test
@@ -111,36 +111,10 @@ configs:
   data_files:
   - split: test
     path: qfrcort/qfrcort.jsonl
-- config_name: daccord
-  data_files:
-  - split: test
-    path: daccord/daccord_test.jsonl
-- config_name: mnli-nineeleven-fr-mt
-  data_files:
-  - split: test
-    path: mnli-nineeleven-fr-mt/multinli_nineeleven_fr_mt_test.jsonl
-- config_name: french_boolq
-  data_files:
-  - split: test
-    path: french_boolq/french_boolq_test.jsonl
-- config_name: rte3-french
-  data_files:
-  - split: validation
-    path: rte3-french/*dev*.jsonl
-  - split: test
-    path: rte3-french/*test*.jsonl
 - config_name: fracas
   data_files:
   - split: test
     path: fracas/fracas_test.jsonl
-- config_name: wino_x_lm
-  data_files:
-  - split: test
-    path: wino_x_lm/wino_x_lm_test.jsonl
-- config_name: wino_x_mt
-  data_files:
-  - split: test
-    path: wino_x_mt/wino_x_mt_test.jsonl
 - config_name: mms
   data_files:
   - split: test
@@ -151,586 +125,134 @@ configs:
     path: wsd/*train*.jsonl
   - split: test
     path: wsd/*test*.jsonl
-pretty_name: Colle
-title: COLE !
-emoji: 🐳
-colorFrom: purple
-colorTo: gray
-sdk: docker
-app_port: 7860
+- config_name: lingnli
+  data_files:
+  - split: train
+    path: lingnli/*train*.jsonl
+  - split: test
+    path: lingnli/*test*.jsonl
+- config_name: multiblimp
+  data_files:
+  - split: train
+    path: multiblimp/*train*.jsonl
+  - split: validation
+    path: multiblimp/*validation*.jsonl
+  - split: test
+    path: multiblimp/*test*.jsonl
+pretty_name: COLE
 ---
 
-# COLE Leaderboard
-Here is our website [leaderboard](https://colebenchmark.org/).
-
-# COLE Dataset Card
-## Dataset Summary
-
-The COLE benchmark is a suit of multiple French NLP tasks for evaluating language models. It includes test sets, and some validation, and training sets for tasks such as sentiment analysis, question answering, NLI, and more.
-
-## Article and Corpus Details
-See our article [COLE: a Comprehensive Benchmark for French Language Understanding Evaluation](https://arxiv.org/abs/2510.05046) for more details about the tasks and the corpus.
-
-## Task Descriptions
-
-## Allocine.fr
-Allo-ciné tests language understanding in sentiment classification by feeding movie reviews which can be either positive and negative, the task consists in giving the correct sentiment for each review.
-
-## DACCORD
-Determine if a French sentence makes sense semantically (binary label).
+# COLE: Comprehensive Benchmark for French Language Understanding Evaluation
 
-## FQuAD
-Fquad is question/answer pair built on high-quality wikipedia articles. The goal of the model in this task is to accurately predict if the answer to the question really can be found in the provided answer.
+[![Paper](https://img.shields.io/badge/Paper-arXiv%3A2510.05046-b31b1b)](https://arxiv.org/abs/2510.05046)
+[![Website](https://img.shields.io/badge/Website-colebenchmark.org-blue)](https://colebenchmark.org/)
+[![GitHub](https://img.shields.io/badge/GitHub-GRAAL--Research%2FCOLE-181717?logo=github)](https://github.com/GRAAL-Research/COLE)
 
-## FraCaS 
-Fracas is a natural language inference (NLI) taskthe where the model must classify the relationship between a premise and a hypothesis-entailment, contradiction, or neutral-based on complex linguistic phenomena such as quantifiers, plurality, anaphora, and ellipsis.
-
-## Fr-BoolQ
-Boolean question answering in French: answer true/false based on context.
-
-## GQNLI-fr
-The dataset consists of carefully constructed premise-hypothesis pairs that involve quantifier logic (e.g. most, at least, more than half). The goal is to evaluate the model's ability to reason about these expressions and determine whether the hypothesis logically follows from the premise, contradicts it, or is neutral.
-
-## LingNLI
-LingNLI is a NLI corpus collected by putting a linguist 'in the loop' to dynamically introduce novel constraints during data collection, aiming to mitigate the systematic gaps and biases often found in crowdsourced datasets.
-
-## MMS-fr 
-MMS-fr is a sentiment analysis task where the model classifies a French text as positive (2), neutral (1), or negative (0), assessing its ability to detect sentiment across diverse domains and sources.
-
-## MNLI-nineeleven-Fr-MT 
-French machine-translated version of MNLI using 9/11 context, for entailment classification.
-
-## MultiBLiMP-Fr 
-MultiBLiMP-Fr is a grammatical judgment task where the model must identify the grammatically correct sentence from a minimal pair differing by a single targeted feature, thereby assessing its knowledge of French syntax, morphology, and agreement.
-
-## PAWS-X
-This task aims to test paraphrase identification by giving two sentences and a label defining if these sentences are equivalent in meaning or not.
-
-## PIAF
-This task consists of pairs of questions and text answers with information of where in the answer is the truly relevant information.
-
-## QFrBLiMP
-This task gives the model sentences pairs, the goal is to determine if the sentences are semantically equivalent, or, put more simply, if they mean the same thing, even with slightly different syntax and words.
-
-## QFrCoLA 
-QFrCoLA is a french dataset made from multiple french language sites such as académie-française.fr and vitrinelinguistique.com. It aims to tests models ability to determine a sentence's acceptability in french on subjects such as grammar and syntax. The answer is a binary label indicating if the sentence is correct or not.
-
-## QFrCoRE
-QFrCoRE is a definition matching task where the model selects the correct standard French  definition for a Quebec French expression from a list of candidates.
-
-## QFrCoRT
-QFrCoRE is a definition matching task where the model selects the correct standard French  definition for a Quebec French term from a list of candidates.
-
-## RTE3-Fr
-French version of RTE3 for textual entailment recognition.
-
-## SICK-fr
-This task also has pairs of sentences and notes them on 2 dimensions, relatedness and entailment. While relatedness scales from 1 to 5, entailement is a choice between entails, contradicts or neutral.
-
-## STS22
-This task evaluates whether pairs of news articles, written in different languages, cover the same story. It focuses on document-level similarity, where systems rate article pairs on a 4-point scale from most to least similar
-
-## Wino-X-LM
-Pronoun resolution task: choose between two referents in a sentence with an ambiguous pronoun.
-
-## Wino-X-MT
-Translation-based pronoun resolution: choose which of two French translations uses the correct gendered pronoun.
-
-## WSD-Fr
-WSD-Fr is a word sense disambiguation task where the model must identify the correct meaning of an ambiguous verb in context, as part of the FLUE benchmark.
-
-## XNLI-fr
-This task consists of pairs of sentences where the goal is to determine the relation between the two sentences, this relation can be either entailement, neutral or contradiction.
+## Dataset Summary
 
+**COLE** is a comprehensive benchmark for evaluating French Natural Language Understanding (NLU). It includes 23 diverse tasks covering sentiment analysis, paraphrase detection, natural language inference, question answering, grammatical judgment, word sense disambiguation, and more — with a particular focus on linguistic phenomena relevant to the French language.
+
+We benchmark 94 large language models (LLMs), providing an extensive analysis of the current state of French NLU. Our results highlight a significant performance gap between closed- and open-weight models and identify key challenging frontiers such as zero-shot extractive question answering, fine-grained word sense disambiguation, and understanding of regional language variations.
+
+For more details, see our paper: [COLE: a Comprehensive Benchmark for French Language Understanding Evaluation](https://arxiv.org/abs/2510.05046) (ICLR 2025 Workshop).
+
+## Links
+
+- **Leaderboard**: [colebenchmark.org](https://colebenchmark.org/)
+- **Paper**: [arXiv:2510.05046](https://arxiv.org/abs/2510.05046)
+- **GitHub**: [GRAAL-Research/COLE](https://github.com/GRAAL-Research/COLE)
+
+## Tasks
+
+COLE consists of 23 tasks grouped by NLU capability:
+
+### Sentiment Analysis
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **Allocine** | Sentiment classification of French movie reviews (positive/negative) | 20,000 |
+| **MMS-fr** | Sentiment analysis with 3 classes (positive, neutral, negative) | 63,190 |
+
+### Natural Language Inference (NLI)
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **FraCaS** | NLI involving quantifiers, plurality, anaphora, and ellipsis | 346 |
+| **GQNLI-fr** | NLI with quantifier logic (e.g., most, at least, more than half) | 30 |
+| **LingNLI** | NLI corpus constructed with a linguist in the loop | 4,893 |
+| **MNLI-nineeleven-Fr-MT** | French machine-translated MNLI using 9/11 context | 2,000 |
+| **RTE3-Fr** | French version of RTE3 for textual entailment | 3,121 |
+| **SICK-fr** | Sentence pair relatedness and entailment | 4,906 |
+| **XNLI-fr** | Cross-lingual NLI in French | 5,010 |
+| **DACCORD** | Semantic plausibility of French sentences (binary) | 1,034 |
+
+### Question Answering
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **FQuAD** | Extractive QA on high-quality French Wikipedia articles | 400 |
+| **Fr-BoolQ** | Boolean question answering in French | 178 |
+| **PIAF** | French extractive QA pairs | 384 |
+
+### Paraphrase Detection
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **PAWS-X** | Paraphrase identification from sentence pairs | 2,000 |
+| **QFrBLiMP** | Semantic equivalence detection between sentence pairs | 2,290 |
+
+### Grammatical Judgment
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **MultiBLiMP-Fr** | Grammatical correctness from minimal pairs | 77 |
+| **QFrCoLA** | Sentence acceptability in French (grammar, syntax) | 7,546 |
+
+### Semantic Similarity
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **STS22** | Document-level similarity of multilingual news articles | 72 |
+
+### Word Sense Disambiguation
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **WSD-Fr** | Disambiguating verb meanings in context | 3,121 |
+
+### Quebec French
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **QFrCoRE** | Matching Quebec French expressions to standard definitions | 4,633 |
+| **QFrCoRT** | Matching Quebec French terms to standard definitions | 201 |
+
+### Coreference / Pronoun Resolution
+| Task | Description | Test size |
+|------|-------------|-----------|
+| **Wino-X-LM** | Pronoun resolution with ambiguous referents | 2,793 |
+| **Wino-X-MT** | Translation-based pronoun resolution with gendered pronouns | 2,988 |
 
 ## Language
-The language data in COLE is in French .
-
-### Dataset structure
-
-## Allocine.fr
-```json
 
-{
-  "review": "Magnifique épopée, une belle histoire, touchante avec des acteurs qui interprètent très bien leur rôles (Mel Gibson, Heath Ledger, Jason Isaacs...), le genre de film qui se savoure en famille! :)",
-  "label": 1
-}
-```
-
-## DACCORD
-```json
-
-{
-  "id": "a001",
-  "premise": "Le camion-remorque de la vidéo transporte un long tube cylindrique, qui est une pièce destinée à une raffinerie de pétrole en Ouzbékistan.",
-  "hypothesis": "Le camion-remorque de la vidéo transporte un missile nucléaire russe.",
-  "label": "1",
-  "label_text": "contradiction",
-  "url": "https://factuel.afp.com/doc.afp.com.32MJ3M7-1",
-  "genre": "conflit ukrainien-russe"
-}
-```
-
-## FQuAD
-```json
-
-{
-  "title": "pégase_23_3",
-  "context": "D'anciennes théories associent Pégase au combat naval, ou voient en lui un simple navire...",
-  "question": "Quand le théologien Jacques-Paul Migne s'exprime au sujet de Méduse ?",
-  "answers": {
-    "answers_start": [509, 512, 512],
-    "text": ["en 1855", "1855", "1855"]
-  },
-  "is_impossible": false
-}
-```
-
-## FraCaS
-```json
+All data in COLE is in **French**.
 
-{
-  "id": "1",
-  "premise": "Un Italien est devenu le plus grand ténor du monde.",
-  "hypothesis": "Il y a eu un Italien qui est devenu le plus grand ténor du monde.",
-  "label": "0",
-  "question": "Y a-t-il eu un Italien qui soit devenu le plus grand ténor du monde ?",
-  "answer": "yes",
-  "premises_original": "An Italian became the world's greatest tenor.",
-  "premise1": "Un Italien est devenu le plus grand ténor du monde.",
-  "premise1_original": "An Italian became the world's greatest tenor.",
-  "premise2": "",
-  "premise2_original": "",
-  "premise3": "",
-  "premise3_original": "",
-  "premise4": "",
-  "premise4_original": "",
-  "premise5": "",
-  "premise5_original": "",
-  "hypothesis_original": "There was an Italian who became the world's greatest tenor.",
-  "question_original": "Was there an Italian who became the world's greatest tenor?",
-  "note": "",
-  "topic": "GENERALIZED QUANTIFIERS"
-}
-```
-
-## Fr-BoolQ
-```json
+## Usage
 
-{
-  "question": "Jobs avait-il des sautes d'humeur inimaginables durant la période où il dirigeait NeXT ?",
-  "passage": "Il a beaucoup été question de la personnalité agressive et exigeante de Steve Jobs. [...] Dan’l Lewin, déclare dans ce même magazine que Steve Jobs, durant cette période, « avait des sautes d'humeur inimaginables » [...]",
-  "label": 1
-}
-```
+```python
+from datasets import load_dataset
 
-## GQNLI-Fr
-```json
+# Load a specific task
+dataset = load_dataset("graalul/COLE-public", "allocine")
 
-{
-  "uid": 214,
-  "premise": "Il y a six ours. Trois ours marron, un ours noir et un ours blanc courent le long de l'herbe cyan.",
-  "hypothesis": "Un ours beige court.",
-  "label": 1,
-  "label_text": "neutral",
-  "premise_original": "There are six bears...",
-  "hypothesis_original": "One beige bear runs."
-}
+# Available configs: allocine, daccord, fquad, fracas, french_boolq,
+# gqnli, lingnli, mms, mnli-nineeleven-fr-mt, multiblimp, paws_x,
+# piaf, qfrblimp, qfrcola, qfrcore, qfrcort, rte3-french, sickfr,
+# sts22, wino_x_lm, wino_x_mt, wsd, xnli
 ```
 
-## LingNLI
-```json
-
-{
-  "premise": "La richesse des citations verbatim - constituant un bon tiers de ce livre - améliore également la vraisemblance de Burn Rate.",
-  "hypothesis": "Burn Rate manque de véracité et n'inclut aucune référence à d'autres œuvres d'aucune sorte.",
-  "label": 2
-}
-```
-
-## MMS
-```json
-
-{
-  "text": "Cadeaux pour ma fille.",
-  "label": 2
-}
-```
-
-## MNLI-nineeleven-Fr-MT
-```json
-
-{
-  "premise": "La faillite du nationalisme laïque et autocratique était évidente dans le monde musulman à la fin des années 1970.",
-  "hypothesis": "Les musulmans détestaient le nationalisme autocratique à la fin des années 1970.",
-  "label": "1",
-  "label_text": "neutral",
-  "pairID": "62534e",
-  "promptID": "62534",
-  "premise_original": "The bankruptcy of secular, autocratic nationalism was evident across the Muslim world by the late 1970s.",
-  "hypothesis_original": "Muslims disliked autocratic nationalism by the late 1970s."
-}
-```
-
-## MultiBLiMP-Fr
-```json
-
-{
-  "sentence_a": "C'est le genre à lequel appartiennent les espèces de kiwi.",
-  "sentence_b": "C'est le genre à lequel appartenez les espèces de kiwi.",
-  "label": 0
-}
-```
-
-## PAWS-X
-```json
-
-{
-  "id": 12,
-  "sentence1": "La rivière Tabaci est un affluent de la rivière Leurda en Roumanie.",
-  "sentence2": "La rivière Leurda est un affluent de la rivière Tabaci en Roumanie.",
-  "label": 0
-}
-```
-
-## PIAF
-```json
-
-{
-  "id": "p140295203922856",
-  "title": "Alaungpaya",
-  "context": "Il ne convainquit cependant pas tout le monde. Après la chute d'Ava le 23 mars 1752, son propre père lui conseilla de se soumettre : il lui fit valoir que, bien qu'ayant des quantités de soldats enthousiastes, il manquait de mousquets et que leur petite palissade ne résisterait jamais à une armée bien équipée qui venait de mettre à sac Ava, puissamment fortifiée. Alaungpaya, impavide, déclara : « Quand on combat pour son pays, il importe peu qu'on soit rares ou nombreux. ce qui compte est que vos camarades aient un cœur sincère et des bras forts. » Il prépara sa défense en fortifiant Moksobo (renommé Shwebo), avec une palissade et des douves. Il fit couper la forêt à l'extérieur, détruire les mares et combler les puits.",
-  "question": "De quoi Alaungpaya aurait il eu besoin pour remporter la bataille ?",
-  "answers": {
-    "text": ["de mousquets et que leur petite palissade"],
-    "answer_start": [222]
-  }
-}
-```
-
-## QFrBLiMP
-```json
-
-{
-  "id": 250,
-  "label": 0,
-  "ungrammatical": "Cette femme chante très haute.",
-  "source": "https://vitrinelinguistique.oqlf.gouv.qc.ca/...",
-  "category": "morphology",
-  "type": 11,
-  "subcat": 13.0,
-  "grammatical": "Cette femme chante très haut.",
-  "options": [
-    {"id": "1", "text": "La phrase numéro 1"},
-    {"id": "2", "text": "La phrase numéro 2"}
-  ],
-  "answer": "accept"
-}
-```
-
-## QFrCoLA
-```json
-
-{
-  "label": 1,
-  "sentence": "Je vous en prie, soyez bref.",
-  "source": "https://vitrinelinguistique.oqlf.gouv.qc.ca/...",
-  "category": "anglicism"
-}
-```
-
-## QFrCoRE
-```json
-
-{
-  "expression": "Avoir la chienne",
-  "choices": [
-    "Prendre une chaise et s'asseoir.",
-    "Avoir du plaisir, parfois avec une connotation sexuelle.",
-    "Prépare-toi, ça va brasser.",
-    "Tomber amoureux.",
-    "Être en pleine forme.",
-    "Critiquer sévèrement.",
-    "Personne inefficace, qui ne travaille pas bien.",
-    "Il se comporte mal en public.",
-    "Se détendre, arrêter de s'énerver.",
-    "Avoir peur."
-  ],
-  "correct_index": 9,
-  "reference": "https://canada-media.ca/expressions-quebecoises/"
-}
-```
-
-## QFrCoRT
-```json
-{
-  "terme": "Adonner",
-  "choices": [
-    "se payer du bon temps",
-    "tu sais",
-    "Voici quelques éléments typiques pour décrire l'hiver québécois :La bordée de neige(tempête de neige) de la fin décembre nous a laissédes bancs de neige(congères) sur le bord des rues. Nous avons eu quelques épisodes depoudrerie(blizzard) qui ont rendu les déplacements difficiles, surtout en voiture. Mais c'est vraimentla glace noire(verglas) qui cause le plus d'accidents. Il faudra attendre jusqu'auredoux(remontée des températures) pour que la neige et la glace se transforment ensloche(gadoue constituée de neige fondante et d'eau) puis disparaissent au retour du printemps.",
-    "En hiver, il ne faut pas s'encabaner!Ce joli verbe vient du nom \" cabane \" qui désigne un petit espace de rangement. S'encabaner, c'est donc \" rester dans sa cabane (sa maison), ne pas sortir, rester cloitré chez soi \". Mais comme le disaient les membres du groupe Mes Aïeux dans leur chanson \" Dégénération \" : \" Il ne faut pas rester encabané! \" (surtout en hiver).",
-    "On ne parle pas ici de la neige de la veille.Cette expression signifie \" avoir de l'expérience,voir venir les choses \".",
-    "Ça n'a étrangement absolument rien à voir avec le fait qu'il manque quelque chose. Ben manque se veut plutôt un synonyme de \" peut-être \" , \" sûrement \" ou \" probablement \" . Particulièrement utilisé du côté nord de la Gaspésie et sur la pointe, ben manque que tu risques de l'entendre si tu te promènes dans ces coins-là.",
-    "gant de toilette",
-    "Unpoisson d'avrilest une plaisanterie que l'on fait le 1er avril à une connaissance.",
-    "avoir de la monnaie",
-    "Le verbe \" adonner \" s'utilise pour parler de quelque chose qui se produit de façon fortuite, d'une coïncidence. Il peut avoir différentes nuances de sens selon le contexte.Exemples : \" Tu vas à Québec cette fin de semaine? Ça adonne que moi aussi! Faisons du covoiturage! \"\" Je devais commencer mes cours de zumba ce soir mais ça adonne mal : mon fils est malade! \""
-  ],
-  "correct_index": 9,
-  "reference": "https://vivreenfrancais.mcgill.ca/capsules-linguistiques/expressions-quebecoises/"
-}
-```
-
-## RTE3-Fr
-```json
-
-{
-  "id": "1",
-  "language": "fr",
-  "premise": "La vente a été faite pour payer la facture fiscale de 27,5 milliards de dollars de Yukos, Yuganskneftegaz a été vendu à l'origine pour 9,4 milliards de dollars à une entreprise peu connue, Baikalfinansgroup, qui a ensuite été rachetée par la compagnie pétrolière publique russe Rosneft.",
-  "hypothesis": "Baikalfinansgroup a été vendu à Rosneft.",
-  "label": "0",
-  "label_text": "entailment",
-  "task": "IE",
-  "length": "short"
-}
-```
-
-## SICK-Fr
-```json
-
-{
-  "Unnamed: 0": 5,
-  "label": 2,
-  "relatedness_score": 3.2999999523,
-  "sentence_A": "Deux chiens se battent et se câlinent.",
-  "sentence_B": "Il n'y a pas de lutte et d'étreinte de chiens."
-}
-```
-
-## STS22
-```json
-
-{
-  "id": "1559147599_1558534688",
-  "score": 1.0,
-  "sentence1": "KABYLIE (TAMURT) – Les répercussions économiques...",
-  "sentence2": "Le décret n° 2020-293 du 23 mars 2020..."
-}
-```
-
-## Wino-X-LM
-```json
-
-{
-  "qID": "3UDTAB6HH8D37OQL3O6F3GXEEOF09Z-1",
-  "sentence": "The woman looked for a different vase for the bouquet because it was too small.",
-  "context_en": "The woman looked for a different vase for the bouquet because _ was too small.",
-  "context_fr": "La femme a cherché un vase différent pour le bouquet car _ était trop petit.",
-  "option1_en": "the bouquet",
-  "option2_en": "the vase",
-  "option1_fr": "le bouquet",
-  "option2_fr": "le vase",
-  "answer": 2,
-  "context_referent_of_option1_fr": "bouquet",
-  "context_referent_of_option2_fr": "vase"
-}
-```
-
-## Wino-X-MT
-```json
-
-{
-  "qID": "3FULMHZ7OUVKJ7S9R0LMS753751M44-1",
-  "sentence": "As the wolf approached the house, the man quickly took the knife and not the gun to defend himself because it was near him.",
-  "translation1": "Alors que le loup s'approchait de la maison, l'homme prit rapidement le couteau et non l'arme pour se défendre car il était près de lui.",
-  "translation2": "Alors que le loup s'approchait de la maison, l'homme prit rapidement le couteau et non l'arme pour se défendre car elle était près de lui.",
-  "answer": 1,
-  "pronoun1": "il",
-  "pronoun2": "elle",
-  "referent1_en": "knife",
-  "referent2_en": "gun",
-  "true_translation_referent_of_pronoun1_fr": "couteau",
-  "true_translation_referent_of_pronoun2_fr": "arme",
-  "false_translation_referent_of_pronoun1_fr": "couteau",
-  "false_translation_referent_of_pronoun2_fr": "arme"
-}
-```
-
-## WSD-Fr
-```json
-
-{
-  "sentence": "Il rend hommage au roi de France et des négociations aboutissent au traité du Goulet , formalisant la paix entre les deux pays .",
-  "labels_idx": [10],
-  "label": "négociations"
-}
-```
-
-## XNLI-Fr
-```json
-
-{
-  "premise": "Ils m'ont dit qu'à la fin, on m'amènerait un homme pour que je le rencontre.",
-  "hypothesis": "Le gars arriva un peu en retard.",
-  "label": 1
-}
-```
-
-## Allocine.fr
-| split      | # examples |
-|------------|-----------:|
-| train      |             |
-| validation |       20,000 |
-| test       |       20,000 |
-
-## DACCORD
-| split      | # examples |
-|------------|-----------:|
-| test       |       1,034 |
-
-## FQuAD
-| split      | # examples |
-|------------|-----------:|
-| validation |       100 |
-| test       |       400 |
-
-## FraCaS
-| split      | # examples |
-|------------|-----------:|
-| test       |       346 |
-
-## Fr-BoolQ
-| split      | # examples |
-|------------|-----------:|
-| test       |        178 |
-
-## GQNLI-Fr
-| split      | # examples |
-|------------|-----------:|
-| train      |       243 |
-| validation |        27 |
-| test       |        30 |
-
-## LingNLI
-| split      | # examples |
-|------------|-----------:|
-| train      |     29,985 |
-| test       |      4,893 |
-
-## MMS
-| split      | # examples   |
-|------------|-------------:|
-| train      |      132,696 |
-| validation |       14,745 |
-| test       |       63,190 |
-
-## MNLI-nineeleven-Fr-MT
-| split      | # examples |
-|------------|-----------:|
-| test       |       2,000 |
-
-## MultiBLiMP-Fr
-| split      | # examples |
-|------------|-----------:|
-| train      |        160 |
-| validation |         18 |
-| test       |         77 |
-
-## PAWS-X
-| split      | # examples |
-|------------|-----------:|
-| train      |     49,401  |
-| validation |       2,000 |
-| test       |       2,000 |
-
-## PIAF
-| split      | # examples |
-|------------|-----------:|
-| train      |       3,105 |
-| validation |        346 |
-| test       |        384 |
-
-## QFrBLiMP
-| split      | # examples |
-|------------|-----------:|
-| train      |       NA   |
-| validation |       2,061 |
-| test       |       2,290 |
-
-## QFrCoLA
-| split      | # examples |
-|------------|-----------:|
-| train      |      15,846 |
-| validation |       1,761 |
-| test       |       7,546 |
-
-## QFrCoRE
-| split      | # examples |
-|------------|-----------:|
-| test       |       4,633 |
-
-## QFrCoRT
-| split      | # examples |
-|------------|-----------:|
-| test       |        201 |
-
-## rte3-Fr
-| split      | # examples   |
-|------------|-------------:|
-| train      |      269,821 |
-| validation |         800 |
-| test       |        3,121 |
-
-## SICK-fr
-| split      | # examples |
-|------------|-----------:|
-| train      |       4,439 |
-| validation |        495 |
-| test       |       4,906 |
-
-## STS22
-| split      | # examples |
-|------------|-----------:|
-| train      |        101 |
-| test       |         72 |
-
-## Wino-X-LM
-| split      | # examples |
-|------------|-----------:|
-| test       |       2,793 |
-
-## Wino-X-MT
-| split      | # examples |
-|------------|-----------:|
-| test       |       2,988 |
-
-## WSD
-| split      | # examples |
-|------------|-----------:|
-| test       |       3,121 |
-| train       |       269,821  |
-
-## XNLI-Fr
-| split      | # examples   |
-|------------|-------------:|
-| train      |      393,000 |
-| validation |        2,490 |
-| test       |        5,010 |
-
-
-## Citation 
+## Citation
 
 ```bibtex
 @misc{beauchemin2025colecomprehensivebenchmarkfrench,
-      title={COLE: a Comprehensive Benchmark for French Language Understanding Evaluation}, 
+      title={COLE: a Comprehensive Benchmark for French Language Understanding Evaluation},
       author={David Beauchemin and Yan Tremblay and Mohamed Amine Youssef and Richard Khoury},
       year={2025},
       eprint={2510.05046},
       archivePrefix={arXiv},
       primaryClass={cs.CL},
-      url={https://arxiv.org/abs/2510.05046}, 
+      url={https://arxiv.org/abs/2510.05046},
 }
 ```
-