kepom/Qwen2.5-Coder-1.5B-Instruct-abliterated
kepom/Qwen2.5-Coder-1.5B-Instruct-abliterated is a 1.5 billion parameter instruction-tuned causal language model, derived from Qwen2.5-Coder-1.5B-Instruct and uncensored using the abliteration technique. With a 32768-token context length, this model is optimized for coding tasks and offers improved performance in IF_Eval benchmarks compared to its base model. It is designed for developers seeking a less restricted coding assistant.
Loading preview...
Model Overview
kepom/Qwen2.5-Coder-1.5B-Instruct-abliterated is a 1.5 billion parameter instruction-tuned model based on the Qwen2.5-Coder-1.5B-Instruct architecture. This version has been modified using the "abliteration" technique to remove censorship, aiming to provide a more open-ended coding assistant. It maintains a substantial context length of 32768 tokens, making it suitable for handling extensive code snippets and complex programming instructions.
Key Differentiators
- Uncensored Version: This model is an abliterated variant of the original Qwen2.5-Coder-1.5B-Instruct, offering a less restricted response generation. The abliteration technique is detailed in this article.
- Improved IF_Eval Performance: Benchmarks indicate a notable improvement in IF_Eval scores (45.41 vs 43.43) compared to the base model, suggesting enhanced capabilities in certain coding-related evaluations.
- Coder-Focused: As part of the Qwen2.5-Coder series, it is inherently designed and optimized for code generation, understanding, and related programming tasks.
Usage and Availability
This model can be easily integrated into applications using the Hugging Face transformers library. It is also available for use with Ollama under huihui_ai/qwen2.5-coder-abliterate:1.5b. The model is part of a family of abliterated Qwen2.5-Coder models, ranging from 0.5B to 32B parameters.