richardyoung/Agents-A1-4B-heretic
The richardyoung/Agents-A1-4B-heretic is a 4.5 billion parameter language model, based on InternScience's Agents-A1-4B, which is built on Qwen3.5. This model has been 'abliterated' using the Heretic v2.0.0.dev0 project to significantly reduce refusals (by 89%) while maintaining original capabilities, making it suitable for unrestricted local inference. It is specifically tuned for tool use, multi-step tasks, and agentic reasoning across domains like long-horizon search, engineering, and scientific research.
Loading preview...
Model Overview
richardyoung/Agents-A1-4B-heretic is a 4.5 billion parameter model derived from InternScience's Agents-A1-4B, which is based on Qwen3.5. This version has undergone a process called 'abliteration' using the Heretic project to remove the 'refusal direction' from the model's residual stream. This surgical modification disables safety refusals without retraining, resulting in an 89% reduction in refusals compared to the original model.
Key Capabilities
- Decensored Output: Significantly reduced refusals (11/100 vs. 99/100 for the original) while maintaining core capabilities, as indicated by a very low KL divergence (0.060).
- Agentic Reasoning: Excels at decomposing complex tasks, planning, and adapting strategies.
- Tool Use: Natively supports function calling and integration with external tools like APIs, code interpreters, and search engines.
- Scientific & Professional Reasoning: Capable of tool-integrated scientific reasoning and professional knowledge question answering.
- Instruction Following: Precisely follows detailed, multi-constraint instructions across diverse domains.
Good For
- Research: Exploring model behavior without safety constraints.
- Creative Writing: Unrestricted content generation.
- Education: Studying alignment techniques and model modifications.
- Unrestricted Local Inference: Deploying for use cases requiring full model compliance.