data-is-better-together - GitLite

KO | EN

gitlite — search

#javascript #python #llm #android #machine-learning #deep-learning #linux #docker #windows #awesome-list #security #pytorch #ios #mcp

data-is-better-together

(★ 275)

Let's build better datasets, together!

파일 탐색기

최종 버전 다운로드 (.zip)

- - update-fineweb-progres.yml
- dependabot.yml
- - 00_imgsys_shuffled_deduplicated_cleaned.py
  - 01_synthetic_data_generation_images.py
  - 01_synthetic_data_generation_prompts.py
  - 01_synthetic_data_generation_total.py
  - 02_image_prefernces_cleaned_filtered_sfw.py
  - 03_upload_to_argilla.ipynb
  - 04_binarize_preference_results.ipynb
  - 05_fine_tune_flux_lora.ipynb
  - README.md
  - requirements.txt
  - template.html
- - README.md
- - 01_setup_prompt_translation_space.ipynb
  - 02_upload_prompt_translation_data.ipynb
  - 03_create_dashboard.ipynb
  - README.md
  - requirements.in
  - Translation_with_distilabel_gpt_4_turbo.ipynb
- - README.md
- - 00_datasets_exploration.ipynb
  - 01_data_prep.ipynb
  - 02_load_from_argilla.ipynb
  - aya_dpo_gen.py
  - custom_preference_to_argilla.py
  - instructions.md
  - README.md
  - requirements.in
  - requirements.txt
- - 01_create_preference_task.ipynb
  - preference_gen.py
  - README.md
  - requirements.in
  - requirements.txt
.gitignore
README.md

# 설치 가이드

1. 코드 내려받기

git clone https://github.com/huggingface/data-is-better-together

깃허브에서 프로젝트 코드 전체를 내 컴퓨터로 내려받습니다.

cd data-is-better-together

방금 내려받은 프로젝트 폴더 안으로 이동합니다.

2. Python

쉬움 추천

사전 준비물

Git GitHub에서 프로젝트 코드를 내려받으려면 필요합니다.
Python 3 설치 시 'Add Python to PATH' 옵션을 꼭 체크하세요 (Windows).

⚠️ 이 프로젝트는 규모가 큰 저장소라, 이 방법이 실제 핵심 제품이 아니라 내부 하위 패키지를 가리키는 것일 수 있습니다. README 전체를 함께 확인해보세요.

pip install -r community-efforts/image_preferences/requirements.txt

requirements.txt 등에 명시된 파이썬 라이브러리를 설치합니다.

jupyter notebook

브라우저에서 노트북(.ipynb) 파일들을 열람하고 실행할 수 있는 Jupyter 화면을 켭니다.

✅ 에러 메시지 없이 실행되고 터미널에 안내 문구가 출력되면 정상입니다.

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

<p align="center">
  <img src="https://huggingface.co/blog/assets/community-datasets/thumbnail.png" width="500px"/>
</p>

<p align="center">🤗 <a href="https://huggingface.co/data-is-better-together" target="_blank">Spaces & Datasets</a></p>

# Data is Better Together

> If you are working on a valuable community-developed dataset but are limited by available resources, please reach out to us on the Hugging Face discord. We may be able to provide support to enhance your project.

Data is Better Together is a collaboration between 🤗 Hugging Face, 🏓 Argilla, and the Open-Source ML community. We aim to empower the open-source community to build impactful datasets collectively. This initiative consists of two main components: the community efforts and the cookbook efforts.

<details open>
  <summary><strong>Community Efforts</strong>: They were guided by the HF Team, hands-on projects focused on creating valuable datasets. These projects required the participation of the community and have been successfully completed.</summary>

<ul>
  <details>
  <summary><strong>Prompt ranking</strong></summary>

- **Goal**: This project aimed to create a dataset of 10k prompts ranked by quality. These prompts included both synthetic and human-generated from various datasets. The intention was to use the final dataset for prompt ranking tasks or synthetic data generation. You can find more information about this project in the [prompt ranking README](community-efforts/prompt_ranking/README.md)
  - **How**: First, we prepared a dataset with the prompts to be ranked using Argilla in a Hugging Face Space. Then, we invited the community to rank the prompts based on their quality. Finally, we collected the annotations and released the dataset.
  - **Result**: Over 385 people joined this initiative! Thanks to their contribution, we released [data-is-better-together/10k_prompts_ranked](https://huggingface.co/datasets/data-is-better-together/10k_prompts_ranked). This dataset can be used for different tasks as you can filter the higher-quality prompts (for instance, see the MPEP project) and generate the corresponding completions. You can also find some models built on top of it [here](https://huggingface.co/models?dataset=dataset:data-is-better-together/10k_prompts_ranked).
  </details>

<details>
  <summary><strong>Multilingual Prompt Evaluation Project (MPEP)</strong></summary>

- **Goal**: There are not enough language-specific benchmarks for open LLMs! So, we wanted to create a leaderboard for more languages by leveraging the community. This way, we could evaluate the performance of models using [AlpacaEval](https://github.com/tatsu-lab/alpaca_eval). You can find more information about this project in the [MPEP README](community-efforts/prompt_translation/README.md).
  - **How**: We selected a subset of 500 high-quality prompts from the [data-is-better-together/10k_prompts_ranked](https://huggingface.co/datasets/data-is-better-together/10k_prompts_ranked) (see the prompt ranking project) and asked the community to help us translate this curated prompt dataset into different languages.
  - **Result**: We achieved to translate the whole dataset for Dutch and Russian, and almost finished with Spanish. Many other languages have also joined this initiative. You can take a look at the resulting datasets [here](https://huggingface.co/datasets?search=MPEP).
  </details>

<details>
  <summary><strong>Image Preferences</strong></summary>

- **Goal**: This project aims to create 10K text to image preference pairs. These pairs can be used to evaluate the performance of image generation models across a wide variety of common image categories, based on prompt with varying levels of difficulty. You can find more information about this project in the [image preferences README](community-efforts/image_preferences/README.md) or within the [blogpost](https://huggingface.co/blog/image-preferences).
  - **How**: We use the prompts from [fal/imgsys-results](https://huggingface.co/datasets/fal/imgsys-results), these prompts are evolved based on complexity and quality for various image categories. We then asked the community to annotate the preference between two generated images for each prompt.
  - **Result**: We achieved to annotate 10K preference pairs. You can take a look at the resulting dataset [here](https://huggingface.co/datasets/data-is-better-together/open-image-preferences-v1-binarized).
  </details>
</ul>

<details open>
  <summary><strong>Cookbook Efforts</strong>: They aim to create guides and tools that help the community in building valuable datasets. They are not guided by the HF team and expected to be handled standalone, allowing you to freely contribute or use them to create your own unique dataset.</summary>

<ul>
  <details>
  <summary><strong>Domain Specific Datasets</strong></summary>

This project aims to bootstrap the creation of more domain-specific datasets for training models. The **goal** is to create a set of tools that help users to collaborate with domain experts. Find out more in the [Domain Specific Datasets README.](cookbook-efforts/domain-specific-datasets/README.md)
  </details>

<details>
  <summary><strong>DPO/ORPO Datasets</strong></summary>

Many languages do not have DPO datasets openly shared on the Hugging Face Hub. The [data-is-better-together/preference_data_by_language](https://huggingface.co/spaces/data-is-better-together/preference_data_by_language) Space gives you an overview of language coverage of DPO datasets for different languages. The **goal** of this project is to help foster a community of people building more DPO-style datasets for different languages. Find out more in this [DPO/ORPO datasets README](cookbook-efforts/dpo-orpo-preference/README.md).
</details>

<details>
  <summary><strong>KTO Datasets</strong></summary>

KTO is another type of preference dataset that can be used to train models to make decisions. Unlike DPO, it doesn't require two candidate responses. Instead, it relies on a simple binary preference, i.e. 👍👎. Thus, data is easier to collect and annotate. The **goal** of this project is to help the community create their own KTO dataset. Find out more in this [KTO datasets README](cookbook-efforts/kto-preference/README.md)

</details>
  </ul>

**🤝 How can I contribute to the cookbook efforts?** That's easy! You can contribute by following the instructions in the README of the project you are interested in. Then, share your results with the community!

</details>