DuoNeural/gemma-4-26B-A4B-it-abliterated-GGUF

Gemma 4 26B-A4B Abliterated — GGUF Q4KM

DuoNeural | BF16 →

GGUF Q4KM of DuoNeural/gemma-4-26B-A4B-it-abliterated. ~16.5GB.

llama.cpp

llama-server -c 131072 \
  -ngl 999 \
  -m gemma-4-26B-A4B-it-abliterated-Q4_K_M.gguf \
  -ctk turbo4 \
  -ctv turbo3 \
  -ub 2048

-ctk turbo4 -ctv turbo3 enables asymmetric KV cache quantization — halves KV VRAM with near-zero perplexity impact.

Ollama

FROM DuoNeural/gemma-4-26B-A4B-it-abliterated-GGUF
PARAMETER num_ctx 32768
PARAMETER temperature 0.7
PARAMETER top_p 0.9

Hardware

• —24GB VRAM (RTX 3090/4090, A40): full offload, 32K context comfortable
• —16GB VRAM: fits with -ctk turbo4 -ctv turbo3, reduced context

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](https://duoneural.beehiiv.com).

DuoNeural Research Publications

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