Huihui-Qwen3.5-27B-abliterated

Upstream

Qwen3.5-27B

Modified by

huihui-ai

Method

Sumandora/remove-refusals-with-transformers

Canonical modified weights

available upstream (high-precision weights published)

Method (as documented): Sumandora/remove-refusals-with-transformers Method parameters: undocumented — no layer range here, unlike candidate 1 which gives 18-51 Modifier: huihui-ai Notes and caveats: - filename_hazard: MALFORMED SHARD NAMES: model.safetensors-00001-of-00011.safetensors ... model.safetensors-00011-of-00011.safetensors (the usual scheme is model-00001-of-00011.safetensors). PRESERVE BYTE-EXACT in any torrent manifest. Naive tooling globbing 'model-*.safetensors' will miss all eleven files. - gguf_note: No huihui-owned GGUF; card points only to Ollama huihui_ai/qwen3.5-abliterated. Alternatives: DevQuasar/huihui-ai.Huihui-Qwen3.5-27B-abliterated-GGUF, cs2764/...-Q6_K-GGUF / -Q8_0-GGUF. Because Qwen3.5-27B is multimodal, VERIFY MMPROJ PRESENCE before endorsing any of these. - adoption_note: Highest-traffic safetensors repo among the Qwen candidates. - ecosystem_warning: huihui also ships huihui-ai/Huihui-Qwen3.5-27B-Claude-4.6-Opus-abliterated, a DIFFERENT model with a 30+ repo derivative tree and confusingly similar downstream names. Also nbeerbower/Huihui-Qwen3.5-27B-abliterated-Athanorlite-ORPO (v1/v2) is a genuine FINETUNE CHILD of this candidate — a real lineage edge if wanted. Source: refusal-modified-acquisition-manifest.json v1.0.0 (research snapshot 2026-09-20), candidate 26.

Explore further