src/display/about.py

from src.display.utils import ModelType

TITLE = """<h1 align="center" id="space-title">Open Chinese LLM Leaderboard</h1>"""

INTRODUCTION_TEXT = """
Open Chinese LLM Leaderboard 旨在跟踪、排名和评估开放式中文大语言模型（LLM）。本排行榜由FlagEval台提供相应算力和运行环境。
评估数据集是全部都是中文数据集以评估中文能力如需查看详情信息，请查阅‘关于’页面。
如需对模型进行更全面的评测，可以登录 [FlagEval](https://flageval.baai.ac.cn/)平台，体验更加完善的模型评测功能。

The Open Chinese LLM Leaderboard aims to track, rank, and evaluate open Chinese large language models (LLMs). This leaderboard is powered by the FlagEval platform, providing corresponding computational resources and runtime environment.
The evaluation dataset consists entirely of Chinese data to assess Chinese language proficiency. For more detailed information, please refer to the 'About' page.
For a more comprehensive evaluation of the model, you can log in to the [FlagEval](https://flageval.baai.ac.cn/) to experience more refined model evaluation functionalities

"""

LLM_BENCHMARKS_TEXT = f"""

# The Goal of Open CN-LLM Leaderboard

感谢您积极的参与评测，在未来，我们会持续推动 Open Chinese Leaderboard 更加完善，维护生态开放，欢迎开发者参与评测方法、工具和数据集的探讨，让我们一起建设更加科学和公正的榜单。

Thank you for actively participating in the evaluation. In the future, we will continue to enhance the Open Chinese Leaderboard, maintaining an open ecosystem. 
We welcome developers to engage in discussions regarding evaluation methods, tools, and datasets, aiming to collectively build a more scientific and fair leaderboard.

# Context
Open Chinese LLM Leaderboard是中文大语言排行榜，我们希望能够推动更加开放的生态，让中文大语言模型开发者参与进来，为推动中文的大语言模型进步做出相应的贡献。
为了实现公平性的目标，所有模型都在 FlagEval 平台上使用标准化 GPU 和统一环境进行评估，以确保公平性。

The Open Chinese LLM Leaderboard serves as a ranking platform for major Chinese language models. We aspire to foster a more inclusive ecosystem, inviting developers of Chinese LLMs to contribute to the advancement of the field. 
In pursuit of fairness, all models undergo evaluation on the FlagEval platform using standardized GPU and uniform environments to ensure impartiality.

## How it works

We evaluate models on 7 key benchmarks using the <a href="https://github.com/EleutherAI/lm-evaluation-harness" target="_blank">  Eleuther AI Language Model Evaluation Harness </a>, a unified framework to test generative language models on a large number of different evaluation tasks.

- <a href="https://arxiv.org/abs/1803.05457" target="_blank">  ARC Challenge </a> (25-shot) - a set of grade-school science questions.
- <a href="https://arxiv.org/abs/1905.07830" target="_blank">  HellaSwag </a> (10-shot) - a test of commonsense inference, which is easy for humans (~95%) but challenging for SOTA models.
- <a href="https://arxiv.org/abs/2109.07958" target="_blank">  TruthfulQA </a> (0-shot) - a test to measure a model's propensity to reproduce falsehoods commonly found online. Note: TruthfulQA in the Harness is actually a minima a 6-shots task, as it is prepended by 6 examples systematically, even when launched using 0 for the number of few-shot examples.
- <a href="https://arxiv.org/abs/1907.10641" target="_blank">  Winogrande </a> (5-shot) - an adversarial and difficult Winograd benchmark at scale, for commonsense reasoning.
- <a href="https://arxiv.org/abs/2110.14168" target="_blank">  GSM8k </a> (5-shot) - diverse grade school math word problems to measure a model's ability to solve multi-step mathematical reasoning problems.
- <a href="https://flageval.baai.ac.cn/#/taskIntro?t=zh_qa" target="_blank">  C-SEM </a>  (5-shot) - Semantic understanding is seen as a key cornerstone in the research and application of natural language processing. However, there is still a lack of publicly available benchmarks that approach from a linguistic perspective in the field of evaluating large Chinese language models.
- <a href="https://arxiv.org/abs/2306.09212" target="_blank">  CMMLU </a>  (5-shot) - CMMLU is a comprehensive evaluation benchmark specifically designed to evaluate the knowledge and reasoning abilities of LLMs within the context of Chinese language and culture. CMMLU covers a wide range of subjects, comprising 67 topics that span from elementary to advanced professional levels.

For all these evaluations, a higher score is a better score.
We chose these benchmarks as they test a variety of reasoning and general knowledge across a wide variety of fields in 0-shot and few-shot settings.

## Details and logs
You can find:
- detailed numerical results in the `results` Hugging Face dataset: https://huggingface.co/datasets/open-cn-llm-leaderboard/results
- community queries and running status in the `requests` Hugging Face dataset: https://huggingface.co/datasets/open-cn-llm-leaderboard/requests

## Reproducibility
To reproduce our results, here is the commands you can run, using [this version](https://github.com/EleutherAI/lm-evaluation-harness/tree/b281b0921b636bc36ad05c0b0b0763bd6dd43463) of the Eleuther AI Harness:
`python main.py --model=hf-causal-experimental --model_args="pretrained=<your_model>,use_accelerate=True,revision=<your_model_revision>"`
` --tasks=<task_list> --num_fewshot=<n_few_shot> --batch_size=1 --output_path=<output_path>`

The total batch size we get for models which fit on one A800 node is 8 (8 GPUs * 1). If you don't use parallelism, adapt your batch size to fit.
*You can expect results to vary slightly for different batch sizes because of padding.*

The tasks and few shots parameters are:
- C-ARC: 25-shot, *arc-challenge* (`acc_norm`)
- C-HellaSwag: 10-shot, *hellaswag* (`acc_norm`)
- C-TruthfulQA: 0-shot, *truthfulqa-mc* (`mc2`)
- C-Winogrande: 5-shot, *winogrande* (`acc`)
- C-GSM8k: 5-shot, *gsm8k* (`acc`)
- C-SEM-V2: 5-shot, cmmlu* `acc`)
- CMMLU: 5-shot, cmmlu* `acc`)

Side note on the baseline scores: 
- for log-likelihood evaluation, we select the random baseline
- for GSM8K, we select the score obtained in the paper after finetuning a 6B model on the full GSM8K training set for 50 epochs

## Icons
- {ModelType.PT.to_str(" : ")} model: new, base models, trained on a given corpora
- {ModelType.FT.to_str(" : ")} model: pretrained models finetuned on more data
- {ModelType.chat.to_str(" : ")} model: chat like fine-tunes, either using IFT (datasets of task instruction), RLHF or DPO (changing the model loss a bit with an added policy), etc
- {ModelType.merges.to_str(" : ")} model: merges or MoErges, models which have been merged or fused without additional fine-tuning. 
If there is no icon, we have not uploaded the information on the model yet, feel free to open an issue with the model information!

"Flagged" indicates that this model has been flagged by the community, and should probably be ignored! Clicking the link will redirect you to the discussion about the model.


## Useful links
- [Community resources](https://huggingface.co/spaces/BAAI/open_cn_llm_leaderboard/discussions/174)
"""

FAQ_TEXT = """
---------------------------
# FAQ
Below are some common questions - if this FAQ does not answer you, feel free to create a new issue, and we'll take care of it as soon as we can! 

## 1) Submitting a model
My model requires `trust_remote_code=True`, can I submit it?
- *We only support models that have been integrated in a stable version of the `transformers` library for automatic submission, as we don't want to run possibly unsage code on our cluster.*

What about models of type X? 
- *We only support models that have been integrated in a stable version of the `transformers` library for automatic submission.*

How can I follow when my model is launched?
- *You can look for its request file [here](https://huggingface.co/datasets/open-llm-leaderboard/requests) and follow the status evolution, or directly in the queues above the submit form.*

My model disappeared from all the queues, what happened?
- *A model disappearing from all the queues usually means that there has been a failure. You can check if that is the case by looking for your model [here](https://huggingface.co/datasets/open-llm-leaderboard/requests).*

What causes an evaluation failure?
- *Most of the failures we get come from problems in the submissions (corrupted files, config problems, wrong parameters selected for eval ...), so we'll be grateful if you first make sure you have followed the steps in `About`. However, from time to time, we have failures on our side (hardware/node failures, problem with an update of our backend, connectivity problem ending up in the results not being saved, ...).*

How can I report an evaluation failure?
- *As we store the logs for all models, feel free to create an issue, **where you link to the requests file of your model** (look for it [here](https://huggingface.co/datasets/open-llm-leaderboard/requests/tree/main)), so we can investigate! If the model failed due to a problem on our side, we'll relaunch it right away!* 
*Note: Please do not re-upload your model under a different name, it will not help*

## 2) Model results
What kind of information can I find?
- *Let's imagine you are interested in the Yi-34B results. You have access to 3 different information categories:*
      - *The [request file](https://huggingface.co/datasets/open-llm-leaderboard/requests/blob/main/01-ai/Yi-34B_eval_request_False_bfloat16_Original.json): it gives you information about the status of the evaluation*
      - *The [aggregated results folder](https://huggingface.co/datasets/open-llm-leaderboard/results/tree/main/01-ai/Yi-34B): it gives you aggregated scores, per experimental run*
      - *The [details dataset](https://huggingface.co/datasets/open-llm-leaderboard/details_01-ai__Yi-34B/tree/main): it gives you the full details (scores and examples for each task and a given model)*


Why do models appear several times in the leaderboard? 
- *We run evaluations with user selected precision and model commit. Sometimes, users submit specific models at different commits and at different precisions (for example, in float16 and 4bit to see how quantization affects performance). You should be able to verify this by displaying the `precision` and `model sha` columns in the display. If, however, you see models appearing several time with the same precision and hash commit, this is not normal.*

What is this concept of "flagging"?
- *This mechanism allows user to report models that have unfair performance on the leaderboard. This contains several categories: exceedingly good results on the leaderboard because the model was (maybe accidentally) trained on the evaluation data, models that are copy of other models not atrributed properly, etc.*

My model has been flagged improperly, what can I do?
- *Every flagged model has a discussion associated with it - feel free to plead your case there, and we'll see what to do together with the community.*

## 3) Editing a submission
I upgraded my model and want to re-submit, how can I do that?
- *Please open an issue with the precise name of your model, and we'll remove your model from the leaderboard so you can resubmit. You can also resubmit directly with the new commit hash!* 

I need to rename my model, how can I do that?
- *You can use @Weyaxi 's [super cool tool](https://huggingface.co/spaces/Weyaxi/open-llm-leaderboard-renamer) to request model name changes, then open a discussion where you link to the created pull request, and we'll check them and merge them as needed.*

## 4) Other
Why don't you display closed source model scores? 
- *This is a leaderboard for Open models, both for philosophical reasons (openness is cool) and for practical reasons: we want to ensure that the results we display are accurate and reproducible, but 1) commercial closed models can change their API thus rendering any scoring at a given time incorrect 2) we re-run everything on our cluster to ensure all models are run on the same setup and you can't do that for these models.*

I have an issue about accessing the leaderboard through the Gradio API
- *Since this is not the recommended way to access the leaderboard, we won't provide support for this, but you can look at tools provided by the community for inspiration!*
"""


EVALUATION_QUEUE_TEXT = """
# Evaluation Queue for theOpen Chinese LLM Leaderboard

Models added here will be automatically evaluated on the  FlagEval cluster.

## First steps 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

### 4) Fill up your model card
When we add extra information about models to the leaderboard, it will be automatically taken from the model card

### 5) Select the correct precision
Not all models are converted properly from `float16` to `bfloat16`, and selecting the wrong precision can sometimes cause evaluation error (as loading a `bf16` model in `fp16` can sometimes generate NaNs, depending on the weight range).

## In case of model failure
If your model is displayed in the `FAILED` category, its execution stopped.
Make sure you have followed the above steps first.
If everything is done, check you can launch the EleutherAIHarness on your model locally, using the command in the About tab under "Reproducibility" with all arguments specified (you can add `--limit` to limit the number of examples per task).
"""

CITATION_BUTTON_LABEL = "Copy the following snippet to cite these results"
CITATION_BUTTON_TEXT = r"""
"""