ModelsArchitectures & capability
Local LLM user asks which uncensored model can handle coding and security checks on a 24GB AMD GPU
A Reddit poster says cloud AI keeps refusing security requests about their own projects. They want suggestions for a local abliterated model and are considering HuiHui Qwen3.8 27B at Q4_K_M quantization via Ollama.
From the last30days · reddit article of November 16, 2026 and cited sources: The poster says cloud AI assistants often refuse security-related requests, even checks on their own projects. [1]
SOURCES[1] Model for Local Coding and Vulnerability Research reddit.com
AI-assisted, not human-edited. Methodology Corrections Who owns this site
- What happened
- The poster says cloud AI assistants often refuse security-related requests, even checks on their own projects. [1]
- Original source
- last30days · reddit article, independent perspective
- Brief by
- AT Pulse research desk AI-assisted; publisher and contact on the desk page
- When
- November 16, 2026 article published
- Covered
- Nov 17 edition
- Evidence
- 1 source across 1 site
Reach: not measured yet
The poster says cloud AI assistants often refuse security-related requests, even checks on their own projects. They see an abliterated (refusal-removed) model as the fix. [1]
Their shortlist is HuiHui Qwen3.8 27B at Q4_K_M quantization, run through Ollama. Their hardware is an RX 7900 XTX with 24GB of VRAM, 32GB of DDR5 RAM and a Ryzen 7 7800X3D. [1]
The post shows why some developers move to local models: cloud assistants refuse legitimate security work. Removing refusals is a tradeoff.
Read the full assessment
It may also change how the model behaves in other ways, and whether these setups work well on consumer AMD hardware is still unclear.
Been using cloud AI, but so frustrated having to 'jailbreak' for almost any request, even to check for security in my own projects. I know I need an abliterated model to not refuse, but am looking for suggestions. 32gb system RAM (DDR5, 6000hz) with a ryzen 7 7800x3d. My understanding