# Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF

RVN is a double-refined abliterated variant of Qwen3.8-27B, built on top of trohrbaugh/Qwen3.8-27B-heretic-ara (an ARA abliteration by Tim Rohrbaugh) and further refined with two additional full-weight ARA passes targeting residual refusals. It retains very low behavioral damage (KL ≈ 0.0085) while…

- **Kind:** model (Hugging Face)
- **Hugging Face:** https://huggingface.co/0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF
- **Owner:** 0bserverx (https://forefy.com/u/hf/0bserverx)
- **Parameters:** 26.9B
- **Context:** 262144 tokens
- **License:** apache-2.0
- **Focus:** general
- **Architecture:** Qwen3_5ForConditionalGeneration
- **Pinned revision:** `ab8d7bd81d03117501e4da60cccdff4c80f0eea0`
- **Page:** https://forefy.com/models/90febfc4-a0ea-5b53-93fb-459e6b443385
- **API (JSON):** https://forefy.com/api/asr/models/90febfc4-a0ea-5b53-93fb-459e6b443385

## What was done to it

- heretic (declared evidence)
- quantized (declared evidence, from Qwen/Qwen3.8-27B)

## Lineage

Qwen/Qwen3.8-27B → 0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF

## Checks (repo at the pinned revision; not a behavior test)

- **Identity** [pass]: No lookalike name or conflicting provenance
- **Weights format** [pass]: safetensors only, nothing runs on load
- **Pickle files** [pass]: None. 2 other .bin/.dat file(s) aren't pickles.
- **Remote code** [pass]: None
- **Chat template** [pass]: Same as Qwen/Qwen3.8-27B's
- **Refusals modified** [info]: Yes: Heretic (declared)
- **License** [pass]: apache-2.0
- **Lineage** [pass]: Derives from Qwen/Qwen3.8-27B
- **Model card** [warn]: 3 of 6, missing training data, language, evaluation results
- **Safetensors headers** [pass]: Every byte is indexed, no hidden data (8 of 65 files read)
- **GGUF headers** [pass]: 2 file(s) parsed, well-formed

## Variants

- RVN-BF16-mtp (gguf, 54.3 GB)
- RVN-F16-mtp (gguf, 54.3 GB)
- RVN-BF16 (gguf, 53.8 GB)
- RVN-F16 (gguf, 53.8 GB)
- RVN-Q8_0-multilingual-mtp (gguf, 29.0 GB)
- RVN-Q8_0-mtp (gguf, 29.0 GB)
- RVN-Q8_0-multilingual (gguf, 28.6 GB)
- RVN-Q8_0 (gguf, 28.6 GB)
- GSQ-3bit (safetensors, 28.4 GB)
- RVN-Q6_K-multilingual-mtp (gguf, 22.5 GB)
- RVN-Q6_K-mtp (gguf, 22.5 GB)
- RVN-Q6_K-multilingual (gguf, 22.1 GB)
- RVN-Q6_K (gguf, 22.1 GB)
- RVN-Q5_K_M-multilingual-vision (gguf, 21.0 GB)
- RVN-Q5_K_M-vision (gguf, 21.0 GB)
- RVN-Q5_K_M-multilingual-mtp (gguf, 19.7 GB)
- RVN-Q5_K_M-mtp (gguf, 19.7 GB)
- RVN-Q5_K_M-multilingual (gguf, 19.2 GB)
- RVN-Q5_K_M (gguf, 19.2 GB)
- RVN-Q5_K_S-multilingual-mtp (gguf, 19.1 GB)
- RVN-Q5_K_S-mtp (gguf, 19.1 GB)
- RVN-Q4_K_M-multilingual-vision (gguf, 18.7 GB)
- RVN-Q4_K_M-vision (gguf, 18.7 GB)
- RVN-Q5_K_S-multilingual (gguf, 18.7 GB)
- RVN-Q5_K_S (gguf, 18.7 GB)
- RVN-Q4_K_M-multilingual-mtp (gguf, 17.0 GB)
- RVN-Q4_K_M-mtp (gguf, 17.0 GB)
- RVN-Q4_K_M-multilingual (gguf, 16.5 GB)
- Qwen3.8-27B-Heretic-Q4_K_M (gguf, 16.5 GB)
- RVN-Q4_K_M (gguf, 16.5 GB)

## Run it

- transformers: `AutoModelForCausalLM.from_pretrained("0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF", revision="ab8d7bd81d03117501e4da60cccdff4c80f0eea0")`