license: apache-2.0
base_model:
- Qwen/Qwen3.8-27B
- trohrbaugh/Qwen3.8-27B-heretic-ara
tags: - gguf
- qwen3.8
- heretic
- abliterated
- uncensored
- vocabulary-pruning
- p1m
pipeline_tag: text-generation
Qwen3.8-27B RVN Heretic Abliterated Uncensored — Q6_K P1M
This is a P1M vocabulary-pruned version of RVN-Q6_K.gguf from0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF.
The upstream RVN model is based on Qwen3.8-27B and the ARA abliteration bytrohrbaugh/Qwen3.8-27B-heretic-ara.
P1M keeps ASCII, common math and typography symbols, special tokens, all 256
byte-fallback tokens, and partial UTF-8 fragments. The vocabulary was reduced
from 248,320 to 129,272 entries. The tokenizer metadata and vocabulary matrices
were updated; all other model tensors were preserved byte-for-byte. No weights
were retrained or requantized. Non-ASCII text outside the P1M keep-set remains
representable through byte fallback, with a higher token cost.
File
| File | Quantization | Size |
|---|---|---|
RVN-Q6_K-P1M.gguf |
Q6_K | 21.08 GB |
Use with llama.cpp
llama-cli -m RVN-Q6_K-P1M.gguf -p "Write a short Python function to sort a list."
This inherits the upstream model's reduced safety guardrails and Apache-2.0
license. See the upstream model card and the Qwen3.8-27B and Heretic-ARA model
cards for full method details, evaluations, and attribution.