This repository is a curated list of open-source projects and resources related to DeepSeek Coder, a family of large language models specialized for code generation and understanding. It serves developers and researchers looking to leverage or build upon these powerful coding models.

How It Works

The list categorizes projects into official DeepSeek Coder models (available in various sizes from 1.3B to 33B parameters, with base and instruct variants), community-built models that fine-tune or adapt DeepSeek Coder, and quantized versions (AWQ, GGUF, GPTQ) for easier deployment. It highlights integrations with popular AI coding assistants like Tabby, AutoDev, and Copilot refact.

Quick Start & Requirements

• Models: Available on Hugging Face (e.g., huggingface.co/deepseek-ai).
• Quantized Models: Provided by TheBloke in AWQ/GGUF/GPTQ formats.
• Integrations: Projects like Tabby offer self-hosted AI coding assistance, with DeepSeek Coder 6.7B noted as a top performer on their leaderboard. AutoDev provides an IDE AI assistant with DeepSeek 6.7B fine-tuning tools.

Highlighted Details

• DeepSeek Coder 6.7B is highlighted as the top performer for code completion on the Tabby leaderboard.
• Offers models in base and instruct variants across multiple parameter sizes (1.3B, 5.7B, 6.7B, 33B).
• Includes quantized versions (AWQ, GGUF, GPTQ) for broader accessibility and deployment flexibility.
• Showcases community projects like OpenCodeInterpreter and Magicoder built upon DeepSeek Coder.

Maintenance & Community

• The list is curated by deepseek-ai, the creators of DeepSeek Coder.
• Community contributions are encouraged for new projects and resources.

Licensing & Compatibility

• The specific license for DeepSeek Coder models is not detailed in this list, but models are generally available on Hugging Face. Users should verify individual model licenses.
• Quantized models by TheBloke typically have permissive licenses, but verification is recommended.

Limitations & Caveats

• This is a curated list of related projects, not the DeepSeek Coder models themselves. Users must navigate to individual project links for setup and usage details.
• Some model variants (e.g., 5.7B instruct) are listed as "coming soon."