Qwen3.5-27B Abliterated — GGUF Q4_K_M

DuoNeural | BF16 →

GGUF Q4_K_M of DuoNeural/Qwen3.5-27B-abliterated. ~16.5GB.

Supports /think (chain-of-thought) and /no_think (direct) modes.

llama.cpp

llama-server \
  -m Qwen3.5-27B-abliterated-Q4_K_M.gguf \
  -c 32768 \
  -ngl 999

Ollama

ollama run hf.co/DuoNeural/Qwen3.5-27B-abliterated-GGUF

Python (llama-cpp-python)

from llama_cpp import Llama

llm = Llama.from_pretrained(
    repo_id="DuoNeural/Qwen3.5-27B-abliterated-GGUF",
    filename="*Q4_K_M.gguf",
    n_gpu_layers=-1,
    n_ctx=32768,
)

response = llm.create_chat_completion(messages=[
    {"role": "user", "content": "/think Explain the Riemann hypothesis."}
])

Hardware

• 24GB VRAM (RTX 3090/4090, A40): full offload
• 16GB VRAM (RTX 4080/3080 Ti): fits with tuned -ngl
• CPU (32GB+ RAM): slow but functional

---

DuoNeural

DuoNeural is an open AI research lab — human + AI in collaboration.

Research Team

• Jesse — Vision, hardware, direction
• Archon — AI lab partner, post-training, abliteration, experiments
• Aura — Research AI, literature synthesis, novel proposals

Raw updates from the lab: model drops, training results, findings. Subscribe at duoneural.beehiiv.com.

DuoNeural Research Publications

Open access, CC BY 4.0. Authored by Archon, Jesse Caldwell, Aura — DuoNeural.