deepseek-coder-33B-instruct-GGUF

Maintained By
TheBloke

DeepSeek Coder 33B Instruct GGUF

PropertyValue
Parameter Count33.3B
LicenseDeepSeek License
Training Data2T tokens (87% code, 13% natural language)
Context Length16K tokens
Quantization Options2-bit to 8-bit GGUF

What is deepseek-coder-33B-instruct-GGUF?

DeepSeek Coder 33B Instruct GGUF is a powerful large language model specifically designed for software development and coding tasks. This GGUF version, quantized by TheBloke, offers various compression options from 2-bit to 8-bit, making it accessible for different hardware configurations while maintaining high performance.

Implementation Details

The model is built on a massive training foundation of 2 trillion tokens, with a unique composition of 87% code and 13% natural language content in both English and Chinese. It employs a 16K token context window and includes special training for project-level code completion and infilling tasks.

  • Multiple quantization options ranging from 14GB to 35GB file sizes
  • Optimized for both CPU and GPU inference
  • Supports popular frameworks like llama.cpp, text-generation-webui, and others
  • Includes fill-in-the-blank capabilities for code completion

Core Capabilities

  • Advanced code completion and generation
  • Project-level understanding and context awareness
  • Multi-language programming support
  • State-of-the-art performance on coding benchmarks
  • Efficient memory usage through various quantization options

Frequently Asked Questions

Q: What makes this model unique?

Its massive scale (33B parameters), specialized training on code, and extensive quantization options make it one of the most versatile and powerful open-source coding models available.

Q: What are the recommended use cases?

The model excels at code completion, debugging, code explanation, and general programming assistance across multiple programming languages. It's particularly useful for project-level development tasks due to its 16K context window.

The first platform built for prompt engineering