--- task_categories: - text-generation language: - it size_categories: - 1K [!NOTE] > We will release the evaluation script soon. Stay tuned! The evaluation is based on the `lm-evaluation-harness` library, which provides a simple and flexible way to evaluate LLMs on a wide range of tasks and datasets. The tasks are defined in `tasks/medschool-entrance-exams`. To run the evaluation, you can use the following command: ```bash MODEL_ARGS="pretrained=meta-llama/Meta-Llama-3.1-8B-Instruct,dtype=bfloat16" lm_eval \ --model hf \ --model_args $MODEL_ARGS \ --tasks medschool_entrance_exams_it_mc,medschool_entrance_exams_it_cloze \ --batch_size auto \ --log_samples \ --output_path outputs/ \ --include tasks/medschool-entrance-exams/ ``` This command evaluates the model `meta-llama/Meta-Llama-3.1-8B-Instruct` on the Italian version of the dataset in both multiple-choice and cloze-style formats. The evaluation results are saved in the `outputs/` directory. Please, refer to the [lm-evaluation-harness](https://github.com/eleutherai/lm-evaluation-harness) repository for more details on how to use the library. ## Data ### Source The dataset is collected from the official Italian website of the Ministry of Education, University and Research ([MIUR](https://www.miur.gov.it/)), which hosts a large collection of past entrance exams for medical school in Italy. The dataset includes questions from various subjects, such as biology, chemistry, physics, mathematics, world knowledge, and more. You can find the original dataset [here](https://domande-ap.mur.gov.it/domande). ### Composition The dataset contains over 3K questions, making it suitable for training and evaluating LLMs. The following table shows the number of questions in the dataset for each subject: | Subject | Number of questions | |---------------|---------------------| | Biology | 1,180 | | Chemistry | 1,009 | | Math & Physics | 655 | | World knowledge | 245 | | Logic & Reasoning | 212 | | **Total** | **3,301** | ### Size comparison with other benchmarks The dataset contains over 3K questions, making it suitable for training and evaluating LLMs. The following table shows the size of the dataset compared to other benchmarks: | **Dataset** | **Number of questions** | |--------------------------|------------------------| | Medical School Entrance Exams | ~3,300 | | ARC-Challenge (test) | ~1,170 | | ARC-Easy (test) | ~2,380 | | BoolQ (validation) | ~3,270 | | CommonSenseQA (validation) | ~1,220 | | GSM8K (test) | ~1,320 | | MMLU (test) | ~14,000 | | PIQA (validation) | ~1,850 | | SciQ (test) | ~1,000 | | TruthfulQA (test) | ~820 | | WinoGrande (test) | ~1,767 | ### Format The dataset is provided in JSONL format, with each line representing a single question in the following format: ```json { "id": 1691, "topic": "biologia", "text": "Come sono definite le cellule staminali che sono in grado di differenziarsi in tutti i tipi di cellule presenti nel corpo umano, ma non possono dare origine ad un organismo completo?", "answers": [ "Cellule Staminali Multipotenti", "Cellule Staminali Pluripotenti", "Cellule Staminali Totipotenti", "Cellule Staminali Unipotenti", "Cellule Staminali Oligopotenti" ], "label": 1 } ``` where: - `id` (int): The unique identifier of the question. - `topic` (str): The subject of the question. - `text` (str): The text of the question. - `answers` (list): A list of five answer choices, with one correct answer. - `label` (int): The index of the correct answer in the `answers` list (0-indexed). ### Examples #### Biology ```json { "id": 1691, "topic": "biologia", "text": "Come sono definite le cellule staminali che sono in grado di differenziarsi in tutti i tipi di cellule presenti nel corpo umano, ma non possono dare origine ad un organismo completo?", "answers": [ "Cellule Staminali Multipotenti", "Cellule Staminali Pluripotenti", "Cellule Staminali Totipotenti", "Cellule Staminali Unipotenti", "Cellule Staminali Oligopotenti" ], "label": 1 } ``` #### Chemistry ```json { "id": 19, "topic": "chimica", "text": "Rispetto alla classificazione che si trova nella tavola periodica il fluoro fa parte del: ", "answers": [ "gruppo dei gas nobili", "gruppo degli alogeni", "gruppo dei lantanidi", "secondo gruppo", "quarto periodo" ], "label": 1 } ``` #### Math & Physics ```json { "id": 1203, "topic": "fisica-matematica", "text": "Quali sono le coordinate del centro della circonferenza di equazione x^2 + y^2 + 2x – 6y + 5 = 0?", "answers": [ "(2 ; –6)", "(–2 ; 6)", "(1 ; 3)", "(–1 ; 3)", "(2 ; 3)" ], "label": 3 } ``` #### World Knowledge ```json { "id": 11, "topic": "competenze-conoscenze", "text": "Quale dei seguenti è il primo romanzo di Italo Calvino, pubblicato nel 1947?", "answers": [ "Se una notte d'inverno un viaggiatore", "Il barone rampante", "Palomar", "Le cosmicomiche", "Il sentiero dei nidi di ragno" ], "label": 4 } ``` #### Logic & Reasoning ```json { "id": 2539, "topic": "logica", "text": "Tutti i pasticcieri praticano il kendo; Gianluca pratica il kendo. Quale delle seguenti affermazioni aggiuntive consentirebbe di dedurre con certezza che Gianluca è un pasticciere?", "answers": [ "Tra le persone che praticano kendo vi sono dei pasticcieri", "Alcune persone che praticano kendo si chiamano Gianluca", "Alcune persone che praticano kendo sono pasticcieri", "Non è certo che ogni persona che pratica kendo sia anche un pasticciere", "Ogni persona che pratica kendo è anche un pasticciere" ], "label": 4 } ``` ## Contributing Contributions to this dataset are welcome! If you have additional tasks or domains that you would like to include, please submit a pull request. ## License This dataset is released under the [Apache 2.0 License](https://www.apache.org/licenses/LICENSE-2.0). Feel free to use it for research, commercial, or personal purposes.