Huihui Qwen3-Next 80B-A3B (abliterated) GGUF
This is a “derivative, but practical” entry: it’s not a new architecture or fine-tune, but a curated set of GGUF quant files for a large Qwen3-Next instruct model. If your workflow runs LLMs via llama.cpp-compatible runtimes (or tools that build on them), having a prebuilt set of quant levels can be much more convenient than downloading an FP16 checkpoint and converting it yourself.
mradermacher’s card calls out multiple static quant variants (with notes like “fast, recommended” for the Q4_K_S file), and points to third-party guidance on how to choose GGUF quant types. A practical way to start is to download the recommended mid-range quant first, verify the prompt format/behavior against the upstream model, then iterate down or up in size depending on your hardware and quality needs. The model is tagged “abliterated/uncensored,” so it’s worth treating it as a less-safety-constrained variant and using it accordingly.
Quick stats from the listing feed: 2399 downloads.
Source listing: https://huggingface.co/models?sort=modified