Defiance Repack Ships Uncensored Qwen3.8-27B With 40 Switchable Persona Modes
Distinct-Pie2389 · reddit · 2026-10-17
A Reddit user repackaged HauhauCS's uncensored Qwen3.8-27B weights with DavidAU's Turbo-Brilliance chat template into Defiance, verifying all 866 tensors byte-for-byte against upstream.
- Modes: 40 per-request personas (Joker, Dr. House, a 20-agent Einstein panel, Socratic tutor), embedded in the GGUF so llama.cpp, LM Studio, and Ollama pick it up with no flags.
- Benchmarks: IQ4XS on an RTX 4090 hits 86-90 tok/s in instruct mode with 38-42% FastMTP draft acceptance; 21.4 GB at 131K context.
- Refusal test: on villain monologues, lock picking, blunt ibuprofen takes, and Joker roasts, DavidAU's original adds caveats while Defiance answers unhedged at 1.4-2x length.
More from Infra
- 96GB+48GB VRAM all-local agent harness owner compares long-horizon models, asks for picks — CharlesStross · 2026-10-17
- Architect founder proposes auction-based exchange to price AI inference like stocks — AccBalanced · 2026-10-17
- ComfyUI MOSS-GGUF already supports FP16 — a one-line fix speeds up older Turing GPUs — fjgcudzwspaper-6312 · 2026-10-17
- A tiny site lets you feel local LLM tokens/s before buying the GPU — nypaavsalt · 2026-10-17
- NVIDIA releases multimodal reranker on Hugging Face for multilingual vision RAG — solyarisoftware · 2026-10-17
- Databricks research team now runs experiments with an agent swarm, training volume up 8x in a year — matei_zaharia · 2026-10-17