sargurun16/VCoder

TEXT GENERATIONConcurrency Cost:1Model Size:3.1BQuant:BF16Ctx Length:32kTool Calling:SupportedPublished:Jun 15, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Cold

VCoder by sargurun16 is a 3.1 billion parameter, Python-focused coding assistant fine-tuned from Qwen2.5-Coder-3B-Instruct with a 32768 token context length. It was optimized using LoRA and Unsloth on 15,000 Python instruction-response examples from the Python Code Instructions 15K dataset. This model excels at Python code generation, problem-solving, debugging, and algorithm implementation, demonstrating a 7.0% Pass@1 improvement over its base model on HumanEval tasks.

Loading preview...

VCoder: Python-Focused Code Generation Model

VCoder is a 3.1 billion parameter coding assistant developed by sargurun16, fine-tuned from the Qwen2.5-Coder-3B-Instruct base model. It leverages LoRA and Unsloth for efficient adaptation, specifically targeting Python programming tasks. The model was trained on 15,000 instruction-response pairs from the Python Code Instructions 15K dataset, making it highly specialized for Python code generation and related problem-solving.

Key Capabilities

  • Enhanced Python Code Generation: Outperforms its base model on HumanEval, showing a 7.0% Pass@1 improvement.
  • Algorithm and Data Structure Implementation: Capable of assisting with complex algorithm design and data structure problems.
  • Debugging and Refactoring: Aids in identifying and correcting errors, as well as improving existing code.
  • Competitive Programming Support: Useful for coding interview questions and competitive programming scenarios.
  • Function Completion: Efficiently completes Python functions based on given instructions.

Good for

  • Developers primarily working with Python who need a specialized coding assistant.
  • Generating Python code snippets, functions, and solving programming challenges.
  • Debugging and refactoring existing Python codebases.
  • Learning and practicing algorithm design and data structures in Python.
  • Users seeking a compact yet powerful model for Python-specific coding tasks, compatible with GGUF formats for local inference via tools like Ollama, LM Studio, or llama.cpp.

Limitations

  • Primarily optimized for Python; performance may vary for other languages.
  • Benchmark results are based on a subset of HumanEval tasks.
  • May occasionally generate incorrect code, especially for highly specialized or niche domains.

Finetunes
2 models