BlossomsAI/Qwen2.5-Coder-7B-Instruct-Uncensored
BlossomsAI/Qwen2.5-Coder-7B-Instruct-Uncensored is a 7.6 billion parameter instruction-tuned causal language model, developed by BlossomsAI. It is an optimized version of Qwen2.5-Coder-7B-Instruct, specifically modified to remove refusal behaviors using directional intervention techniques. This model is designed for resource-efficient deployment, being fully quantized, and primarily targets use cases requiring a less restrictive code generation and instruction-following model.
Loading preview...
Overview
BlossomsAI/Qwen2.5-Coder-7B-Instruct-Uncensored is a 7.6 billion parameter instruction-tuned model, derived from Qwen2.5-Coder-7B-Instruct. Its primary distinction lies in the removal of refusal behaviors, achieved through directional intervention techniques based on research findings from "Refusal in LLMs is Mediated by a Single Direction". This modification aims to provide a more permissive and direct response generation, particularly useful in coding and instruction-following scenarios where typical LLM refusals might hinder productivity.
Key Capabilities
- Uncensored Instruction Following: Specifically engineered to minimize refusal behaviors, allowing for more direct responses to a wider range of prompts.
- Code Generation: Inherits the code generation capabilities of its base model, Qwen2.5-Coder-7B-Instruct.
- Resource-Efficient Deployment: Available in fully quantized formats (e.g., GGUF) for optimized performance on various hardware.
Good for
- Code Development: Ideal for developers seeking a less restrictive model for generating, debugging, or understanding code across multiple programming languages.
- Instruction-Following Tasks: Suitable for applications where direct and uninhibited responses to instructions are critical, without common LLM refusals.
- Research and Experimentation: Useful for researchers exploring the impact of refusal behavior removal on model utility and safety.