This pack contains work derived from DeepSeek Coder.

    https://github.com/deepseek-ai/DeepSeek-Coder

    Code:   MIT License
    Models: DeepSeek Model License (permits commercial use)
    Paper:  DeepSeek-Coder: When the Large Language Model Meets Programming
            -- The Rise of Code Intelligence, arXiv:2401.14196

Two licenses apply and they are not the same. The repository's CODE is MIT. The
model WEIGHTS are governed by a separate Model License; it does permit
commercial use, but it is not MIT and carries use restrictions. Anything in this
pack that ships weights, or a product built on them, is bound by the Model
License, not by this file.

Derived material:

  * `formats.py` reproduces the fill-in-the-middle sentinel tokens, the instruct
    prompt template (verbatim from `finetune/finetune_deepseekcoder.py`), the
    `<|EOT|>` turn terminator, and the 32014 / 32021 stop-token ids. These are a
    wire protocol -- they must match the upstream tokenizer exactly or the model
    silently misreads the prompt.

  * `order_by_dependency` / `build_repo_context` implement the repo-level
    packing described in the paper's data-creation section (parse intra-repo
    dependencies, order files so dependencies precede dependents, mark each file
    with a `#path` header). The implementation here is ours; the scheme is theirs.

  * `MODELS` records the released sizes and the fact that 5.7b has no instruct
    release, taken from the repository README's model table.

No model weights and no upstream source files are vendored into this pack.

Citation:

    @misc{deepseek-coder,
      author = {Daya Guo and Qihao Zhu and Dejian Yang and Zhenda Xie and Kai Dong
                and Wentao Zhang and Guanting Chen and Xiao Bi and Y. Wu and
                Y.K. Li and Fuli Luo and Yingfei Xiong and Wenfeng Liang},
      title  = {DeepSeek-Coder: When the Large Language Model Meets Programming
                -- The Rise of Code Intelligence},
      journal = {CoRR},
      volume = {abs/2401.14196},
      year   = {2024},
      url    = {https://arxiv.org/abs/2401.14196}
    }
