DeepSeek-Coder.yaml

---
# Thank you for contributing!
# In filling out this yaml file, please follow the criteria as described here:
# https://osai-index.eu/contribute

# You're free to build on this work and reuse the data. It is licensed under CC-BY 4.0, with the
# stipulation that attribution should come in the form of a link to https://osai-index.eu/
# and a citation to the peer-reviewed paper in which the dataset & criteria were published:

# Liesenfeld, A. and Dingemanse, M., 2024. Rethinking open source generative AI: open-washing and the EU AI Act. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (pp. 1774-1787).

# Organization tags:
# - National origin: China
# - Contributor type: Non-academic (Chinese AI Tiger)

# Training compute:
# - Base model training compute: ~1.285e+24 FLOP +- 0.5 OoM (param count & dataset size) [Epoch AI]
# - End model training compute: unknown (likely negligible)

system:
    name: DeepSeek-Coder
    link: https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct-0724
    type: code
    performanceclass: latest
    basemodelname: DeepSeek-V2
    endmodelname: DeepSeek-Coder-V2-Instruct-0724
    endmodellicense: DeepSeek License Agreement
    releasedate: 2024-09
    notes: Continued pretrained from an intermediate checkpoint of DeepSeek-v2; model inheritance is arguable.

org:
    name: DeepSeek
    link: https://www.deepseek.com/
    notes: "DeepSeek (深度求索), founded in 2023, is a Chinese company dedicated to making AGI a reality."

# availability:
datasources_basemodel:
    class: closed
    link: 
    notes: Proprietary dataset.

datasources_endmodel:
    class: closed
    link: https://github.com/deepseek-ai/DeepSeek-Coder-V2/blob/main/paper.pdf
    notes: Data mentioned as consisting of public GitHub repositories and the CommonCrawl. No dataset provided.

weights_basemodel:
    class: open
    link: https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Base
    notes: Weights made available on HuggingFace.

weights_endmodel:
    class: open
    link: https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct-0724
    notes: Weights made available on HuggingFace.

trainingcode:
    class: closed
    link: https://github.com/deepseek-ai/DeepSeek-Coder-V2/
    notes: Repo exists, but does not contain any training code.

# documentation:
code:
    class: closed
    link:  
    notes: No training code, so no documentation.

hardware_architecture:
    class: closed
    link:  
    notes: No hardware architecture information provided.

preprint:
    class: open
    link: https://arxiv.org/pdf/2406.11931
    notes: Preprint made available through arXiv.

paper:
    class: closed
    link: 
    notes: No peer-reviewed paper found.

modelcard:
    class: partial
    link: https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct-0724
    notes: Model card primarily contains inference and licensing information.

datasheet:
    class: closed
    link: 
    notes: No datasheet found.

# access:
licenses:
    class: closed
    link: "https://huggingface.co/deepseek-ai/DeepSeek-Coder-V2-Instruct-0724#6-license"
    notes: "Weights: DeepSeek License Agreement. Code: not released. Data: not disclosed."