Community teams release wave of abliterated Qwen models for 8–24GB local GPUs
udmrzn · x · 2026-11-12
A roundup of newly released abliterated/uncensored Qwen-family models from Huihui, HauhauCS and DavidAU on Hugging Face, with per-VRAM quant recommendations:
- Huihui Qwen3.8 27B Abliterated: billed as one of the strongest local uncensored models, retains vision. 16GB: UD-IQ4XS (14.4GB); 24GB: Q4KM/Q5. Runs via ollama run huihuiai/Qwen3.8-abliterated.
- HauhauCS Qwen3.8 27B Aggressive: deeper de-censoring, near-zero refusals, keeps MTP acceleration and vision. 16GB: IQ4XS (15.7GB); 24GB: Q4/Q5.
- Huihui Qwen3.6 35B-A3B Abliterated MTP: MoE with 3B active params for fast inference. 12–16GB: IQ3/IQ4 with expert offload (32GB+ RAM advised); 24GB: full Q4 in VRAM.
- HauhauCS Qwen3.6 35B-A3B Aggressive: aggressive uncensoring plus high throughput.
Repo links included for each model.
More from Infra
- Together Compiler: agent-driven continual compilation for AI inference — togethercompute · 2026-11-12
- Together Compiler explained: run the model, find bottlenecks, let the model fix them — togethercompute · 2026-11-12
- Kaspersky details Black Owl APT breach of software vendor eyeing supply-chain attack — cyb3rops · 2026-11-12
- Speculative decoding draft models for Qwen3-8B compared: WhiFlash vs Tengyunw benchmarks — maxsloef · 2026-11-12
- AMD runs MolmoAct2 robot policy locally on a Ryzen AI Max+ 395 mini PC — DJiafei · 2026-11-12
- Opinion: AI infra is becoming a coding runtime as intelligence units outgrow tokens — prateekj · 2026-11-12