Overview

This model, saidutta69/Qwen2.5-Coder-3B-Instruct-heretic, is a 3.09 billion parameter instruction-tuned causal language model derived from the Qwen2.5-Coder series by Qwen. It is specifically a decensored version of the original Qwen/Qwen2.5-Coder-3B-Instruct, created using the Heretic v1.2.0 tool. The base Qwen2.5-Coder models are designed for code-specific tasks, building upon the strong Qwen2.5 foundation with 5.5 trillion training tokens, including extensive source code and synthetic data.

Key Capabilities

• Enhanced Code Generation, Reasoning, and Fixing: Significant improvements over previous CodeQwen versions.
• Long Context Window: Supports a full 32,768 tokens, beneficial for complex coding projects.
• Reduced Refusals: This 'heretic' variant shows a dramatically lower refusal rate (3/100) compared to the original model (100/100), indicating less content moderation.
• Foundation for Code Agents: Designed to support real-world applications like Code Agents, while maintaining strengths in mathematics and general competencies.

Good For

• Developers seeking a powerful, smaller-scale code-focused LLM with a long context window.
• Applications requiring code generation, debugging, or reasoning without strict content filtering.
• Experimentation with instruction-tuned models for programming tasks where the original model's refusal rate was a limitation.