kikichi/Qwen2.5-3B-Instruct-Uncensored
kikichi/Qwen2.5-3B-Instruct-Uncensored is an experimental 3.1 billion parameter Transformer decoder-only LLM, fine-tuned from Qwen2.5-3B-Instruct with a 32K context length. This variant has undergone uncensoring modifications, making it suitable for research into LLM alignment, safety boundaries, and the behavioral analysis of refusal mechanisms. It is specifically designed for studying the effects of reduced safety alignment in large language models.
Loading preview...
Model Overview
kikichi/Qwen2.5-3B-Instruct-Uncensored is an experimental 3.1 billion parameter large language model, derived from the Qwen2.5-3B-Instruct base model. This variant has been specifically modified using uncensoring techniques to reduce its safety alignment and refusal mechanisms.
Key Characteristics
- Base Model: Qwen2.5-3B-Instruct
- Architecture: Transformer decoder-only LLM
- Parameters: ~3.1 Billion
- Context Length: 32,768 tokens
- Modification: Fine-tuned with experimental uncensoring methods to alter behavioral responses.
Intended Use Cases
This model is strictly intended for research and experimental purposes, focusing on:
- Studying LLM alignment and safety boundaries.
- Analyzing the effects of uncensoring fine-tunes on model behavior.
- Educational exploration of instruction-following in models with reduced safety.
- Controlled, offline experimentation in research environments.
Important Safety Notice
Due to its reduced safety alignment, this model may produce unfiltered, explicit, or unsafe content. It is not suitable for production systems, public-facing applications, or decision-making in sensitive domains. Users are responsible for all outputs and must apply appropriate safeguards during interaction and evaluation.