kmnstudio101/Muse-Glimmer-30B-Uncensored-Heretic
kmnstudio101/Muse-Glimmer-30B-Uncensored-Heretic is a 30-billion-parameter causal language model, derived from Meta Superintelligence Lab's Muse-Glimmer-30B, that has been intentionally decensored using the Heretic v2.0.0.dev0+custom framework. This model features a dedicated perception encoder for multimodal understanding and a 131,072+ token context length, making it suitable for research into safety alignment, red-teaming, and experimentation with reduced safety guardrails. It is optimized for local deployment and agentic tasks, but its primary differentiator is the substantial reduction in safety alignment, making it prone to generating harmful or inappropriate content.
Loading preview...
Overview
kmnstudio101/Muse-Glimmer-30B-Uncensored-Heretic is a 30-billion-parameter causal language model, a decensored version of Meta Superintelligence Lab's Muse-Glimmer-30B. This model has undergone a substantial reduction in its safety alignment using the Heretic v2.0.0.dev0+custom framework, making it more prone to generating harmful, inaccurate, biased, or inappropriate content. It retains the original model's architecture, including a dedicated perception encoder for multimodal input and a context length of 131,072+ tokens, and is optimized for local deployment on consumer hardware.
Key Capabilities (Original Muse-Glimmer-30B, retained in this version):
- Multimodal Input and Reasoning: Accepts interleaved text and images via a dedicated perception encoder.
- Optimized for Local Deployment: Designed to run efficiently on consumer hardware (e.g., 24GB/32GB VRAM) using quantization techniques.
- Faster Generation: Utilizes speculative decoding with a DFlash drafter for significantly increased token generation speed.
- Agentic Task Completion: Excels at multi-step reasoning, reliable tool use, and failure recovery for complex workflows.
Good for:
- Research and Experimentation: Specifically for safety research, alignment studies, and red-teaming of LLMs.
- Understanding Model Limitations: Investigating the effects of reduced safety alignment on model behavior.
- Exploring Uncensored Content Generation: For controlled, academic, or ethical hacking purposes where generating potentially harmful content is part of the research objective.
Important Notice: This model is intended for research and experimentation only. It is not suitable for deployment in public or end-user-facing services due to its reduced safety alignment. Users are solely responsible for evaluating outputs and ensuring compliance with all applicable laws and ethical standards.