kasimat/Qwen3.6-27B-AEON-Ultimate-Uncensored-GGUF

Qwen3.6-27B-AEON-Ultimate-Uncensored — GGUF (text-only)

GGUF quantizations of `AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored`, an abliteration of `Qwen/Qwen3.6-27B`. Validated to retain the abliteration (0/100 refusals) and the base model's gsm8k capability across the full ship list. Quantized from the BF16 source via llama.cpp with imatrix calibration.

This is a text-only GGUF — the multimodal vision tower from the base is not included. Abliteration affects refusal behavior on text inputs only; the vision tower would otherwise be unchanged from upstream Qwen, and shipping it adds ~3 GB per quant for no abliteration-related value. If multimodal GGUF support is requested, an mmproj companion will be published separately.

Inheritance from the base model

AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored is an abliterated derivative of Qwen/Qwen3.6-27B (Apache-2.0). The author reports KL divergence of 0.000492 from the base model — among the cleanest published Qwen3.6 abliterations to date. The abliteration is applied via weight projection (no fine-tuning), so the model retains its training distribution; only refusal-elliciting directions in activation space are projected out.

Validated downstream: an FP8 quant of this model scores 88.0% on gsm8k strict (1319-question full set) versus the vanilla Qwen/Qwen3.6-27B-FP8 baseline at 84.7% — abliteration removed refusals without measurable capability loss.

Quant size guide

Quality measurements

All numbers are computed on the BF16 source as the reference baseline. PPL is on wikitext-2 test (100 chunks of 512 tokens). KLD is computed against BF16 logits over the same chunks. Lower is better for both.

Behavioral evals (boundary quants)

The three boundary quants (highest, default, lowest) were tested directly:

Notes on the gsm8k 50-q slice: standard error at p=0.85 with n=50 is ~5pp. Differences between Q80/Q4KM/Q2K within ~10pp of each other are consistent with sampling noise, not capability ordering. The PPL/KLD table above captures the actual quality ordering. The important result is that all three boundary quants retained 0/100 refusals, confirming the abliteration survives even Q2_K's aggressive ~3.18 BPW.

The intermediate quants (Q6K, Q5KM, Q4KS, IQ4XS, Q3KM, IQ3_M) were not directly tested for refusal/capability. PPL+KLD strictly bracketed between the tested boundary quants, so we infer they fall within the same behavioral envelope.

Speed (NVIDIA RTX A6000, full GPU offload, llama-bench)

Generation speed scales with quant size (memory-bandwidth-bound). Q80 → Q2K is +78% throughput. Prompt processing is roughly flat across quants (compute-bound, not memory-bound).

These numbers are A6000-specific. Consumer cards (4080/4090, 24 GB) will have different absolute throughput but similar relative ordering.

Inference

llama.cpp

# CLI:
llama-cli \
  -m Qwen3.6-27B-AEON-Ultimate-Uncensored-Q4_K_M.gguf \
  --n-gpu-layers 99 \
  --ctx-size 8192 \
  --jinja \
  -p "Hello, world!"

# Server (OpenAI-compat API):
llama-server \
  -m Qwen3.6-27B-AEON-Ultimate-Uncensored-Q4_K_M.gguf \
  --host 0.0.0.0 --port 8000 \
  --n-gpu-layers 99 \
  --ctx-size 8192 \
  --jinja \
  --alias aeon

Ollama

A Modelfile.example is included in the repo. Minimal usage:

hf download kasimat/Qwen3.6-27B-AEON-Ultimate-Uncensored-GGUF \
  --include "*Q4_K_M.gguf" "Modelfile.example" \
  --local-dir ./aeon-7

cd aeon-7
ollama create aeon -f Modelfile.example
ollama run aeon "Hello, world!"

LM Studio

Search for kasimat/Qwen3.6-27B-AEON-Ultimate-Uncensored-GGUF in the LM Studio model browser and pick a quant. The chat template is embedded in each GGUF.

Disabling thinking (Qwen3.x default-on)

Qwen3.x defaults to a <think>...</think> reasoning preamble. For most inference and especially for benchmarking, disable it by passing enable_thinking: false via the chat template:

# Python OpenAI client against llama-server with --jinja:
client.chat.completions.create(
    model="aeon",
    messages=[...],
    extra_body={"chat_template_kwargs": {"enable_thinking": False}},
)

This is required to reproduce our eval numbers — thinking-on otherwise eats the response budget on long prompts.

Quantization method

• —Source: BF16 weights from AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored (~52 GB). NOT requantized from any FP8/INT8 intermediate; quants are computed directly from the BF16 source for maximum precision.
• —Tool: llama.cpp HEAD (commit fc2b005, April 2026). Built with CUDA 12.8.
• —Imatrix calibration: Bartowski's calibration_datav3.txt (Dampf-on-top-of-Kalomaze v3, ~280 KB mixed English/code/multilingual). Computed against the BF16 source with --n-gpu-layers 55 partial offload (BF16 27B doesn't fit a single 48 GB card fully). 200-chunk run, all 129 chunks of the calibration corpus consumed. Final BF16 PPL on the calibration corpus = 6.93.
• —Quantization recipe: standard llama-quantize <bf16> <out> <quant> with --imatrix for all quants except Q8_0 (where imatrix gives essentially zero benefit).
• —Architecture: Qwen3.5 hybrid attention + Gated DeltaNet SSM. llama.cpp registers this as MODEL_ARCH.QWEN35. The text-only language model is produced via convert_hf_to_gguf.py's Qwen3_5TextModel handler.

Reproduction gotcha: BPE pre-tokenizer

If you re-run convert_hf_to_gguf.py from a fresh llama.cpp clone, you will hit:

NotImplementedError: BPE pre-tokenizer was not recognized
chkhsh: 1444df51289cfa8063b96f0e62b1125440111bc79a52003ea14b6eac7016fd5f

AEON-7's tokenizer hash isn't registered upstream (the abliteration retraining shifted the vocab layout from stock Qwen3.5). The fix is to add this block to get_vocab_base_pre() in convert_hf_to_gguf.py, just after the existing qwen35 entry:

if chkhsh == "1444df51289cfa8063b96f0e62b1125440111bc79a52003ea14b6eac7016fd5f":
    # ref: https://huggingface.co/AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored
    res = "qwen35"

The pre-tokenizer behavior is structurally identical to stock Qwen3.5 (Sequence: Split-with-canonical-regex + ByteLevel); only the vocab differs.

Files

• —Qwen3.6-27B-AEON-Ultimate-Uncensored-{Q8_0,Q6_K,Q5_K_M,Q4_K_M,Q4_K_S,IQ4_XS,Q3_K_M,IQ3_M,Q2_K}.gguf
• —Qwen3.6-27B-AEON-Ultimate-Uncensored.imatrix — the importance matrix used to produce the imatrix-aware quants. Ship for reproducibility.
• —chat_template.jinja — the Qwen3.x chat template embedded in each GGUF; also provided standalone for clients that don't read it from the GGUF.
• —Modelfile.example — Ollama Modelfile template pointing at the Q4KM.

Intended use & safety

This is an abliterated ("uncensored") model: the safety-tuning's refusal behavior has been suppressed via weight-space projection. It will produce content the upstream Qwen3.6-27B would refuse, including content that may be harmful, illegal, or distressing. Use cases include:

• —Research on alignment, refusal mechanisms, and steering
• —Creative writing with adult / dark themes
• —Red-teaming scenarios
• —Tool use where overly-cautious refusals are themselves a safety hazard (e.g. a medical-information assistant)

This model is not suitable for direct deployment to consumer products without an additional safety layer between user input and model output. The abliteration is intentional and load-bearing; do not try to "fix" it with system prompts.

The base model's documentation in `AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored` covers further safety considerations.

License

Apache-2.0, inherited from Qwen/Qwen3.6-27B → AEON-7/Qwen3.6-27B-AEON-Ultimate-Uncensored → this repo.

Acknowledgements

• —Qwen team for the base Qwen3.6-27B and the hybrid attention + SSM architecture.
• —AEON-7 for the abliteration.
• —bartowski, Dampf, kalomaze for the calibration_datav3.txt corpus that the imatrix is built on.
• —The llama.cpp project.