Overview

This model, bfavro73/qwen2.5-coder-7b-pandas-dpo-aligned, is a 7 billion parameter language model fine-tuned from Qwen's Qwen2.5-Coder-7B-Instruct. It leverages offline DPO (Direct Preference Optimization) with a preference dataset specifically curated for Python data analysis, making it highly specialized for such tasks. The model balances representational capacity with runtime requirements, offering a robust solution for developers.

Key Capabilities

• Specialized Code Generation: Enhanced coding capabilities, particularly for Python data analysis.
• Broad Competencies: Maintains strengths in mathematics and general language understanding.
• Extended Context Window: Supports a context window of up to 128K tokens, beneficial for complex coding tasks.
• Real-world Applications: Provides a strong foundation for developing applications like Coding Agents.

Deployment Considerations

• GGUF Format: Available in GGUF format, with Q4_K_M quantization recommended for systems with at least 8GB of RAM.
• Memory Requirements: For CPU inference (GGUF), 32 GB RAM is recommended for smoother operation, while 16 GB is a bare minimum. For GPU inference, 10-12 GB VRAM is suggested for 4-bit quantization, and 16 GB VRAM for FP16 or very long contexts.
• CPU Requirements: A realistic minimum of 8-12 vCPUs is recommended for heavier coding uses with multiple users and long contexts, allowing sufficient threads for the model and application.