| ---
|
| license: mit
|
| language:
|
| - multilingual
|
|
|
| base_model:
|
| - deepseek-ai/deepseek-coder-1.3b-instruct
|
|
|
| tags:
|
| - large-language-model
|
| - code-generation
|
| - instruction-tuned
|
| - multilingual
|
| - programming
|
| - software-development
|
| - efficient-model
|
| - gguf
|
| ---
|
|
|
| # DeepSeek-Coder-1.3B-Instruct
|
|
|
| DeepSeek-Coder-1.3B-Instruct is a compact instruction-tuned code generation model developed by DeepSeek AI, designed for programming assistance, code completion, debugging, reasoning, and software engineering workflows. This repository contains GGUF quantized variants of the model optimized for efficient local inference using llama.cpp.
|
|
|
| Unlike general-purpose language models, DeepSeek-Coder-1.3B-Instruct is specifically optimized for software development tasks. It combines strong instruction-following capability with programming knowledge across multiple languages, enabling efficient code generation, explanation, refactoring, and debugging while maintaining a lightweight deployment footprint.
|
|
|
| The quantized formats significantly reduce memory requirements while preserving coding capability and instruction-following quality, making the model suitable for local development environments, IDE integrations, edge deployments, and resource-efficient programming assistants.
|
|
|
| ---
|
|
|
| # Model Overview
|
|
|
| - **Model Name:** DeepSeek-Coder-1.3B-Instruct
|
| - **Base Model:** deepseek-ai/deepseek-coder-1.3b-instruct
|
| - **Architecture:** Decoder-Only Transformer
|
| - **Parameter Count:** Approximately 1.3 Billion Parameters
|
| - **Modalities:** Text
|
| - **Primary Languages:** Multilingual
|
| - **Developer:** DeepSeek AI
|
| - **License:** MIT
|
|
|
| ---
|
|
|
| # Quantization Formats
|
|
|
| This repository provides various GGUF quantized versions of the DeepSeek-Coder-1.3B-Instruct model optimized for efficient local inference using llama.cpp.
|
|
|
| ## IQ3_M
|
|
|
| - Size reduction of approx 75.03% (641.62 MB) compared to 16-bit (2.51 GB)
|
| - Aggressive 3-bit quantization optimized for lightweight coding assistants and low-memory development environments
|
| - Suitable for code completion, syntax generation, and programming support on consumer hardware
|
| - Enables efficient deployment in local IDEs and resource-constrained systems
|
| - Complex software architecture reasoning and large-scale code generation may experience reduced fidelity compared to higher-precision variants
|
|
|
| ## IQ4_NL
|
|
|
| - Size reduction of approx 70.98% (746.04 MB) compared to 16-bit (2.51 GB)
|
| - Advanced 4-bit non-linear quantization designed to preserve programming accuracy and instruction-following quality
|
| - Better suited for debugging, code explanation, refactoring, and structured code generation workflows
|
| - Designed to reduce quantization loss while maintaining reliable coding performance
|
| - May require slightly increased computational overhead during inference
|
|
|
| ## IQ4_XS
|
|
|
| - Size reduction of approx 72.15% (715.94 MB) compared to 16-bit (2.51 GB)
|
| - Balanced 4-bit quantization focused on efficient inference and dependable software development assistance
|
| - Provides a practical balance between memory efficiency, coding quality, and runtime performance
|
| - Suitable for local coding assistants, educational programming tools, and development automation
|
| - Maintains reliable performance across most practical software engineering workloads
|
|
|
| ---
|
|
|
| # Training Background (Original Model)
|
|
|
| DeepSeek-Coder-1.3B-Instruct is trained with an emphasis on programming languages, software engineering knowledge, instruction following, and code reasoning across diverse code repositories and technical datasets.
|
|
|
| ## Pretraining
|
|
|
| - Large-scale pretraining using source code and technical documentation across multiple programming languages
|
| - Focus on programming semantics, code representation learning, and software engineering knowledge
|
| - Optimized for downstream coding, reasoning, debugging, and code generation tasks
|
|
|
| ## Instruction Tuning
|
|
|
| - Refined using instruction-following datasets focused on software development workflows
|
| - Enhanced for conversational coding assistance, debugging, and structured code generation
|
| - Improved consistency across code completion, explanation, refactoring, and technical question answering
|
|
|
| ---
|
|
|
| # Key Capabilities
|
|
|
| - **Code Generation**
|
| Generates high-quality code across multiple programming languages.
|
|
|
| - **Code Completion**
|
| Assists developers with intelligent code completion and implementation suggestions.
|
|
|
| - **Debugging & Refactoring**
|
| Explains, analyzes, and improves existing source code.
|
|
|
| - **Programming Reasoning**
|
| Supports logical reasoning over algorithms, software design, and implementation details.
|
|
|
| - **Instruction Following**
|
| Produces structured responses for coding and software engineering tasks.
|
|
|
| - **Efficient Local Deployment**
|
| Quantized variants enable practical coding assistants on consumer hardware.
|
|
|
| ---
|
|
|
| # Usage Example
|
|
|
| ## Using llama.cpp
|
|
|
| ```bash
|
| ./llama-cli \
|
| -m SandLogicTechnologies/DeepSeek-Coder-1.3B-Instruct_IQ4_NL.gguf \
|
| -p "Write a Python function that implements Dijkstra's shortest path algorithm."
|
| ```
|
|
|
| ---
|
|
|
| # Recommended Usecases
|
|
|
| - **Code Completion**
|
| Accelerate software development with intelligent code suggestions.
|
|
|
| - **Programming Assistants**
|
| Build lightweight AI coding assistants for local development environments.
|
|
|
| - **Debugging & Code Review**
|
| Analyze, explain, and improve existing source code.
|
|
|
| - **Educational Programming**
|
| Support programming education with code explanations and examples.
|
|
|
| - **Developer Productivity**
|
| Automate repetitive coding tasks and technical documentation generation.
|
|
|
| - **Research & Experimentation**
|
| Evaluate compact code-specialized language models for efficient local inference.
|
|
|
| ---
|
|
|
| # Acknowledgments
|
|
|
| These quantized models are based on the original work by the **DeepSeek AI** development team.
|
|
|
| Special thanks to:
|
|
|
| - The DeepSeek AI team for developing and releasing the DeepSeek-Coder-1.3B-Instruct model.
|
|
|
| - **Georgi Gerganov** and the `llama.cpp` open-source community for enabling efficient quantization and inference via the GGUF format.
|
|
|
| ---
|
|
|
| # Contact
|
|
|
| For questions, feedback, or support, please reach out at **[support@sandlogic.com](mailto:support@sandlogic.com)** or visit **https://www.sandlogic.com/**
|
| |
