Models

← All models

DeepSeek-Coder-Base 1.3B

DeepSeek · 2023-11-02 · 1.3B parameters

DeepSeek-Coder-Base 1.3B is the smallest dense base checkpoint in the original DeepSeek-Coder series, released November 2, 2023. Its artifact-specific model card reports one trillion training tokens and multi-head attention; DeepSeek documents a 16,384-token window, fill-in-the-middle training, and coverage of 87 programming languages for the series. It is the foundation for the corresponding 1.3B instruction-tuned variant.

Benchmark scores