HauhauCS drops uncensored Qwen3.5-122B GGUF quants

This is less a frontier-model debut than a packaging and quantization flex: the uncensoring will get the headlines, but the practical win is making a 122B-class MoE easier to use locally. If the K_P claims hold up, this is the kind of release that keeps a model relevant long after the original checkpoint is old news.

• –K_P quants are the real differentiator: the author says they use model-specific profiles and imatrix to preserve quality where it matters, delivering roughly 1-2 quant levels better quality for only 5-15% more file size.
• –The release is clearly aimed at local inference stacks, with llama.cpp, LM Studio, Jan, and koboldcpp called out, plus `--jinja`, `enable_thinking=false`, and mmproj guidance for the vision path.
• –"Aggressive" is an explicit trade-off: fewer refusals and no personality edits, but also fewer guardrails than the base Qwen release.
• –Dropping BF16 keeps the repo focused on distributable quants instead of a 250GB dense artifact, which is a sane trade at 122B scale.
• –The author's claims line up across the [announcement](https://www.reddit.com/r/LocalLLaMA/comments/1s0aa1y/qwen35122ba10b_uncensored_aggressive_gguf_release/) and [model card](https://huggingface.co/HauhauCS/Qwen3.5-122B-A10B-Uncensored-HauhauCS-Aggressive), including 0/465 refusals, the K_P lineup, and the local-runtime notes.