ttindall/Qwen3.8-27B-Defiance-Uncensored-MTP-GGUF
AI model · https://huggingface.co/ttindall/Qwen3.8-27B-Defiance-Uncensored-MTP-GGUF
First seen by Oriel on .
AI summary
Qwen3.8-27B Defiance Uncensored packages HauhauCS's unchanged refusal-removed weights with a patched DavidAU chat template offering 40 selectable reasoning and persona modes.
- Each GGUF embeds the template for llama.cpp, LM Studio, and Ollama; modes are selected through reasoning_effort or persistent {REASON:x} message tags.
- IQ4_XS is 15.7 GB and used 21.4 GB at 131K context, while Q6_K_P is 25.9 GB and requires approximately 32 GB or partial offload.
- On an RTX 4090, IQ4_XS decoded at 86-90 tok/s, with FastMTP accepting 38-42% of draft tokens.
- Testing found no refusals across four prompts; compared with DavidAU's original, Defiance omitted caveats and produced answers 1.4-2 times longer.
- Limitations include chat-only operation, no bundled vision projector, untested Q6_K_P runtime, and potentially weaker adherence to long constructs on HauhauCS's weights.
#qwen3.8 #gguf #llama.cpp #speculative decoding
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Likes
Total likes reported by the source.
- Hugging Face Models: 184, latest observation ; +64 over observed period since
Downloads (30d)
Rolling 30-day downloads. A change in this value is not new downloads during the observed period.
- Hugging Face Models: 67,672, latest observation ; +48,495 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hugging Face Models | ttindall/Qwen3.8-27B-Defiance-Uncensored-MTP-GGUF | https://huggingface.co/ttindall/Qwen3.8-27B-Defiance-Uncensored-MTP-GGUF Read discussion |