"""
This file contains the text content for the leaderboard client.
"""
HEADER_MARKDOWN = """
# 🇨🇿 BenCzechMark

Welcome to the leaderboard!  
Here you can compare models on tasks in Czech language and/or submit your own model. We use our modified fork of [lm-evaluation-harness](https://github.com/DCGM/lm-evaluation-harness) to evaluate every model under same protocol.

- Head to **Submission** page to learn about submission details.
- See **About** page for brief description of our evaluation protocol & win score mechanism, citation information, and future directions for this benchmark.

"""
LEADERBOARD_TAB_TITLE_MARKDOWN = """
   """

SUBMISSION_TAB_TITLE_MARKDOWN = """
    ## How to submit
    1. Head down to our modified fork of [lm-evaluation-harness](https://github.com/DCGM/lm-evaluation-harness). 
    Follow the instructions and evaluate your model on all 🇨🇿 BenCzechMark tasks, while logging your lm harness outputs into designated folder.

    2. Use our script [compile_log_files.py](https://huggingface.co/spaces/CZLC/BenCzechMark/blob/main/compile_log_files.py) for processing log files from your designated folder into single compact submission file that contains everything we need.  
    Example usage:
   - Download sample outputs for csmpt7b from [csmpt_logdir.zip](https://czechllm.fit.vutbr.cz/csmpt7b/sample_results/csmpt_logdir.zip).
   - Unzip.
   - Run the script as with python (with libs jsonlines and tqdm)
   ```bash 
   python compile_log_files.py \
   -i "<your_local_path_to_folder>/csmpt_logdir/csmpt/eval_csmpt7b*" \
   -o "<your_local_path_to_outfolder>/sample_submission.json"
   ```

    3. Upload your file, and fill the form below!
    
    ## Submission
    To submit your model, please fill in the form below.
    
    - *Team name:* The name of your team, as it will appear on the leaderboard
    - *Model name:* The name of your model
    - *Model type:* The type of your model (chat, pretrained, ensemble)
    - *Parameters (B):* The number of parameters of your model in billions (10⁹)
    - *Input length (# tokens):* The number of input tokens that led to the results
    - *Precision:* The precision with which the results were obtained
    - *Description:* Short description of your submission (optional)
    - *Link to model:* Link to the model's repository or documentation
    - *Upload your results:* Results json file to submit
    
    After filling in the form, click the **Pre-submit model** button. 
    This will run a comparison of your model with the existing leaderboard models. 
    After the tournament is complete, you will be able to submit your model to the leaderboard.
"""

RANKING_AFTER_SUBMISSION_MARKDOWN = """
                    This is how will ranking look like after your submission:
                    """
SUBMISSION_DETAILS_MARKDOWN = """
                    Do you really want to submit a model? This action is irreversible.
                    """
MORE_DETAILS_MARKDOWN = """
Here you can view, how selected model won/lost duels to all other models, in selected 🇨🇿 BenCzechMark category.
"""

MODAL_SUBMIT_MARKDOWN = """
Are you sure you want to submit your model?
"""

ABOUT_MARKDOWN = """
## Abstract
We present **B**en**C**zech**M**ark (BCM), the first multitask and multimetric Czech language benchmark for large language models with a unique scoring system that utilizes the theory of statistical significance. Our benchmark covers 54 challenging, mostly native Czech tasks spanning across 11 categories, including diverse domains such as historical Czech, pupil and language learner essays, and spoken word.

Furthermore, we collect and clean the [BUT-Large Czech Collection](https://huggingface.co/datasets/BUT-FIT/BUT-LCC), the largest publicly available clean Czech language corpus, and continuously pretrain the first Czech-centric 7B language model [CSMPT7B](https://huggingface.co/BUT-FIT/csmpt7b), with Czech-specific tokenization. We use our model as a baseline for comparison with publicly available multilingual models.

## Methodology
While we will reveal more details in our upcoming work, here is how leaderboard ranking works in a nutshell.  

### Prompting Mechanism
Each task (except for tasks from language modelling category) is composed of 5 or more prompts. The performance of every model is then max-pooled over tasks (best performance counts).

### Metrics and Significance Testing
We use the following metrics for following tasks:

- Fixed-class Classification: average area under the curve (one-vs-all average)
- Multichoice Classification: accuracy
- Question Answering: exact match
- Summarization: rouge-raw (2-gram)
- Language Modeling : word-level perplexity

On every task, for every metric we compute test for statistical significance at α=0.05, i.e., the probability that performance model A is equal to the performance model B is estimated to be less then 0.05.
We use the following tests, with varying statistical power:
- accuracy and exact-match: one-tailed paired t-test,
- average area under the curve: bayesian test inspired with (Goutte et al., 2005)[https://link.springer.com/chapter/10.1007/978-3-540-31865-1_25],
- summarization & perplexity: bootstrapping.

### Duel Scoring Mechanism, Win Score
On each task, each model is scored to each model (up to top-50 currently submitted models). For each model, record proportion of won duels: **Win Score**(WS).
Next, the **Category Win Score**(CWS), is computed as an average over model's WSs in that category. Similarly, 🇨🇿 **BenCzechMark Win Score** is computed as model's average CWS across categories. 
The properties of this ranking mechanism include:
- Ranking can change after every submission.
- The across-task aggregation is interpretable: in words, it measures the average proportion of times the model is better.
- It allows utilizing wide spectrum of existing resources, evaluated under different metrics.

## Baseline Setup
The models submitted to leaderboard by the authors were evaluated in following setup:
- max input length: 2048 tokens
- number of shown examples (few-shot mechanism): 3-shot
- truncation: smart truncation
- log-probability aggregation: average-pooling
- chat templates: not used

## Citation
You can use the following citation for this leaderboard and our upcoming work.
```bibtex
@article{fajcik2024benczechmark,
  title = {{B}en{C}zech{M}ark: A Czech-centric Multitask and Multimetric Benchmark for Language Models with Duel Scoring Mechanism},
  author = {Martin Fajcik and Martin Docekal and Jan Dolezal and Karel Ondrej and Karel Benes and Jan Kapsa and Michal Hradis and Zuzana Neverilova and Ales Horak and Michal Stefanik and Adam Jirkovsky and David Adamczyk and Jan Hula and Jan Sedivy and Hynek Kydlicek},
  year = {2024},
  url = {https://huggingface.co/spaces/CZLC/BenCzechMark}
  institution = {Brno University of Technology, Masaryk University, Czech Technical University in Prague, Hugging Face},
}
```


## Authors & Correspondence
- **BenCzechMark Authors & Contributors:**
  - **BUT FIT**
    - Martin Fajčík
    - Martin Dočekal
    - Jan Doležal
    - Karel Ondřej
    - Karel Beneš
    - Jan Kapsa
    - Michal Hradiš
  - **FI MUNI**
    - Zuzana Nevěřilová
    - Aleš Horák
    - Michal Štefánik
  - **CIIRC CTU**
    - Adam Jirkovský
    - David Adamczyk
    - Jan Hůla
    - Jan Šedivý
  - **Hugging Face**
    - Hynek Kydlíček

- **Leaderboard Authors & Contributors:**
    - Jan Doležal - Coding and Troubleshooting
    - Martin Fajčík - Management & Debugging
    - Alexander Polok, Jakub Štetina - Leaderboard Version 0.1


**Correspondence to:**
- Martin Fajčík
- Brno University of Technology, Brno, Czech Republic
- Email: [martin.fajcik@vut.cz](mailto:martin.fajcik@vut.cz)
"""