src/about.py

from dataclasses import dataclass
from enum import Enum


@dataclass
class Task:
    benchmark: str
    metric: str
    col_name: str


# Select your tasks here
# ---------------------------------------------------
class Tasks(Enum):
    # task_key in the json file, metric_key in the json file, name to display in the leaderboard
    task0 = Task("main", "acc", "Overall")


NUM_FEWSHOT = 0  # Change with your few shot
# ---------------------------------------------------


# Your leaderboard name
TITLE = """<h1 align="center" id="space-title">👩‍🏫Invalsi Leaderboard
</h1>"""

"""
<img src="https://huggingface.co/spaces/Crisp-Unimib/INVALSIbenchmark/resolve/main/src/logo-crisp-eng-retina.png" height="800" width="150" style="vertical-align: middle;">
"""

# What does your leaderboard evaluate?
INTRODUCTION_TEXT = """
Welcome into <a href="https://crispresearch.it/"><b>CRISP Bicocca</b></a> Invalsi Leaderboard page! 

We adapted the INVALSI benchmark for automated LLM evaluation, which involves rigorous adaptation of the test format to suit automated processing while retaining the essence of the original tests. In this leaderboard, we provide a detailed assessment of current LLMs, offering a crucial reference point for the academic community.

Researchers are invited to submit their models for ongoing evaluation, ensuring the benchmark remains a current and valuable resource.

For more information on the benchmark, please refer to our arXiv paper <a href="https://arxiv.org/abs/2406.17535"><b>here</b></a> and read the "About" section.
"""

# Which evaluations are you running? how can people reproduce what you have?
LLM_BENCHMARKS_TEXT = f"""
## INVALSI Benchmark 

The INVALSI (National Institute for the Evaluation of the Education and Training System) test has been essential to Italy's educational assessment framework since the 2005-2006 academic year. It aims to evaluate students' competencies across various subjects, including Italian language and mathematics, at multiple stages of the education system in a standardised manner. The test's primary objective is to assess students' linguistic proficiency, a crucial skill that schools are tasked with developing as per national educational guidelines.

Since the test covers a wide range of linguistic and comprehension skills, using it to evaluate LLMs can provide a detailed view of a model's proficiency in handling real-world, nuanced language tasks designed for human learners. The test's structured and standardised nature makes it an excellent benchmark for comparing different LLMs with questions culturally and contextually relevant to Italian speakers. However, our findings are relevant across all languages since they assess various general capabilities, such as word formation and text comprehension abilities. Also, since it is designed for different educational stages, it provides varied levels of complexity and challenge. This aspect can gauge an LLM's capability at different difficulty levels, reflecting its potential scalability and adaptability across simpler to more complex linguistic tasks.

### Data Collection
We have collected from public sources 58 unique tests, divided into 141 unique units, with 2114 questions and 2808 unique items. Some questions are subdivided into multiple items, each requiring a specific answer. 
The data has been gathered from the <a href="https://www.gestinv.it/Index.aspx"><b>Gestinv</b></a> database.
Questions' formatting is sometimes not adequately structured for LLM evaluation; for instance, it is sometimes impossible to automatically transcribe the questions into structured fields, necessitating further inspection of images and PDFs. For this reason, we also collected corresponding PDF files and images. Manual inspection was required to ensure accuracy. In cases where questions involved graphical elements, we modified them into an appropriate multiple-choice format. For example, if the task required a student to find and underscore a word, we reformulated the question to allow selection from multiple choices. Similarly, if the task involved drawing a line between two groups of concepts—a common task for younger students—we rephrased it to involve choosing the correct association from given options. Generally, we aimed to adapt the questions to a format that allows the model to select the correct answer from a pool of choices if it aligns with the original type of question.


### Dataset Characteristics

<ul>
 <li> We have selected <b>10 tests</b> comprising <b>31 unique units</b>, <b>409 questions</b>, and <b>625 items</b> from the above data. A test comprises <b>two or more different units</b>; each question can have more than one item to answer. The sample of tests has been chosen by manual inspection, aiming to include different grades and years and avoiding those with questions that require inspecting an image or contained questions that would be hard to reformulate for language model comprehension. <br> <br>
<table style="font-size: 18px; border-collapse: collapse; width: 60%; margin: auto; border: 1px solid black;">
    <caption style="caption-side: top; text-align: center; font-weight: bold;">
        Distribution of tests, questions and items by educational grade.
    </caption>
    <thead>
        <tr style="background-color: #f2f2f2;">
            <th style="border: 1px solid black; padding: 8px; text-align: left;">School Grade</th>
            <th style="border: 1px solid black; padding: 8px; text-align: left;"># Tests</th>
            <th style="border: 1px solid black; padding: 8px; text-align: left;"># Questions</th>
            <th style="border: 1px solid black; padding: 8px; text-align: left;"># Items</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">2nd Grade (Primary School)</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">2</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">34</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">72</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">5th Grade (Primary School)</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">2</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">75</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">117</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">6th Grade (Middle School)</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">1</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">87</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">118</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">8th Grade (Middle School)</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">2</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">86</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">88</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">10th Grade (High School)</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">2</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">77</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">139</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">13th Grade (High School)</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">1</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">50</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">91</td>
        </tr>
    </tbody>
</table> 
</li>
<br> <br>
<li> Each question in the benchmark is labelled with the specific <b>lexical macro area</b> it aims to assess <br> <br>

<table style="font-size: 18px; border-collapse: collapse; width: 80%; margin: auto; border: 1px solid black;">
    <caption style="caption-side: top; text-align: center; font-weight: bold;">
        Distribution of questions by section and macro area.
    </caption>
    <thead>
        <tr style="background-color: #f2f2f2;">
            <th style="border: 1px solid black; padding: 8px; text-align: left;">Section</th>
            <th style="border: 1px solid black; padding: 8px; text-align: left; width: 3.5cm;">Macro Area</th>
            <th style="border: 1px solid black; padding: 8px; text-align: right;"># Questions</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td rowspan="3" style="border: 1px solid black; padding: 8px; text-align: left; width: 1.3cm;">Text comprehension</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Reconstruct the meaning of the text, locally or globally</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">179 (43.8%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Locate and identify information within the text</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">108 (26.4%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Reflect on the content or form of the text, <br> locally or globally, and evaluate them</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">33 (8.1%)</td>
        </tr>
        <tr>
            <td rowspan="6" style="border: 1px solid black; padding: 8px; text-align: left; width: 1.3cm;">Reflection on the language</td>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Lexicon and semantics</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">29 (7.1%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Morphology</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">24 (5.9%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Syntax</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">19 (4.6%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Word formation</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">7 (1.7%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Textuality and pragmatics</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">6 (1.5%)</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 8px; text-align: left;">Spelling</td>
            <td style="border: 1px solid black; padding: 8px; text-align: right;">4 (1.0%)</td>
        </tr>
    </tbody>
</table>

</li>
<br> <br>

<li> The questions come in five distinct formats <br> <br> 
<table style="width: 90%; margin: auto; border-collapse: collapse; font-size: 18px; border: 1px solid black;">
    <caption style="caption-side: top; text-align: center; font-weight: bold;">
        Combined Legend of Question Types and Distribution by Format
    </caption>
    <thead>
        <tr style="background-color: #f2f2f2;">
            <th style="border: 1px solid black; padding: 10px; text-align: center;">Question Type</th>
            <th style="border: 1px solid black; padding: 10px; text-align: center;">Description</th>
            <th style="border: 1px solid black; padding: 10px; text-align: center;"># Questions</th>
            <th style="border: 1px solid black; padding: 10px; text-align: center;"># Items</th>
        </tr>
    </thead>
    <tbody>
        <tr>
            <td style="border: 1px solid black; padding: 10px; text-align: left;"><i>Multiple Choice (MC)</i></td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">Composed of a question with several answer options, among which only one is correct. The most common question format.</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">334 (81.7%)</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">340</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 10px; text-align: left;"><i>Multiple Complex Choice (MCC)</i></td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">Composed of input questions and multiple items to answer. Requires all items to be correctly answered.</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">38 (9.3%)</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">228</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 10px; text-align: left;"><i>Unique Response (RU)</i></td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">Open-ended questions with only one correct answer, sometimes allowing limited variants.</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">33 (8.1%)</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">50</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 10px; text-align: left;"><i>Short Response (RB)</i></td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">Specific sub-type of RU questions requiring a single identifying word as the answer.</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">1 (0.2%)</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">1</td>
        </tr>
        <tr>
            <td style="border: 1px solid black; padding: 10px; text-align: left;"><i>Cloze (CL)</i></td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">Text with missing words to be filled in, can be open-ended or closed-ended.</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">3 (0.7%)</td>
            <td style="border: 1px solid black; padding: 10px; text-align: left;">6</td>
        </tr>
    </tbody>
</table>
</il>

</ul>


"""

EVALUATION_QUEUE_TEXT = """
## Evaluation Form

Please submit your model for evaluation by filling out the form located [here](https://forms.gle/4d6FYudxCVoTvj1W9).

## Some good practices before submitting a model

### 1) Make sure you can load your model and tokenizer using AutoClasses:
```python
from transformers import AutoConfig, AutoModel, AutoTokenizer
config = AutoConfig.from_pretrained("your model name", revision=revision)
model = AutoModel.from_pretrained("your model name", revision=revision)
tokenizer = AutoTokenizer.from_pretrained("your model name", revision=revision)
```
If this step fails, follow the error messages to debug your model before submitting it. It's likely your model has been improperly uploaded.

Note: make sure your model is public!
Note: if your model needs `use_remote_code=True`, we do not support this option yet but we are working on adding it, stay posted!

### 2) Convert your model weights to [safetensors](https://huggingface.co/docs/safetensors/index)
It's a new format for storing weights which is safer and faster to load and use. It will also allow us to add the number of parameters of your model to the `Extended Viewer`!

### 3) Make sure your model has an open license!
This is a leaderboard for Open LLMs, and we'd love for as many people as possible to know they can use your model 🤗
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = r"""
@misc{mercorio2024disceautdeficereevaluating,
      title={Disce aut Deficere: Evaluating LLMs Proficiency on the INVALSI Italian Benchmark}, 
      author={Fabio Mercorio and Mario Mezzanzanica and Daniele Potertì and Antonio Serino and Andrea Seveso},
      year={2024},
      eprint={2406.17535},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
      url={https://arxiv.org/abs/2406.17535}, 
}
"""

QUESTION_FORMAT_TEXT = """
Performance (accuracy %) comparison of AI models across school grades and question formats for grades 2 to 13.
Grade 2 and 5 are part of primary school, grade 6 and 8 are part of middle school, and grade 10 and 13 are part of high school.
Due to the stratification and the limited number of questions in some categories, extreme values such as 100 or 0 are more attainable in the sections with few items. 

The question formats are abbreviated as:

- _Multiple Choice (MC)_: composed of a question with several answer options, among which only one is correct.
- _Multiple Complex Choice (MCC)_: composed of input questions and multiple items to answer.
- _Unique Response (RU)_: involves open-ended questions in which there are no options or suggestions and where only one answer is considered correct (with sometimes a limited number of possible variants).
- _Short Response (RB)_: is a specific sub-type of RU questions in which the answer is short, tending to consist of a single identifying word.
- _Cloze (CL)_: can be open-ended or closed-ended. In both cases, a text with missing words is provided. The aim is to complete the text by inserting the correct missing terms.
"""

MACRO_AREA_TEXT = """"
Performance (accuracy %) comparison of AI models across semantic macro areas. 
The number of questions for each category is indicated in the table headers. 

Categories are abbreviated as: 

* _LI_: Locate and identify information within the text.
* _RM_: Reconstruct the meaning of the text, locally or globally.
* _RC_: Reflect on the content or form of the text, locally or globally, and evaluate them.
* _WF_: Word formation.
* _LS_: Lexicon and semantics.
* _MO_: Morphology.
* _SP_: Spelling.
* _SY_: Syntax.
* _TP_: Textuality and pragmatics.
"""

CHANGELOG_TEXT = """
## Changelog

### 30/06/2024
- 🆕 New model: google/gemma-2-9b-it

### 28/06/2024
- 🆕 New model: ludocomito/Minerva_3B_Ties_1.0
- 🆕 New model: mii-llm/maestrale-chat-v0.4-beta
- 🆕 New model: Qwen/Qwen2-72B-Instruct
- 🆕 New model: nvidia/Nemotron-4-340B-Instruct

### 27/06/2024
- 📈 Added changelog.
- 🆕 New model: sapienzanlp/modello-italia-9b
- 🆕 New model: mixtral-8x22b-instruct
- 🆕 New model: claude-3.5-sonnet

### 26/06/2024
- 🚀 Initial leaderboard launch.
- 📊 Added evaluation metrics for the INVALSI benchmark.
- 📄 Added leaderboard introduction.
- ✍ Added INVALSI benchmark description.
- 📝 Added evaluation queue form.
- 💬 Added citation button.
- 🔎 Added question format and macro area descriptions.
"""