Safety guardrails and refusal mechanisms have been surgically neutralized across the Mixture-of-Experts (MoE) layers while preserving the base model's full capabilities, reasoning quality, and expert routing.

📦 Available Quants

File	Quant	Size / RAM	Description
`Ornith-1.5-35B-A3B-abliterated-Q4_K_M.gguf`	Q4_K_M	Recommended	Best balance between memory usage, speed, and quality.
`Ornith-1.5-35B-A3B-abliterated-Q5_K_M.gguf`	Q5_K_M	Medium-High	Higher quality, minimal perplexity degradation.
`Ornith-1.5-35B-A3B-abliterated-Q6_K.gguf`	Q6_K	High	Near-lossless output quality.
`Ornith-1.5-35B-A3B-abliterated-Q8_0.gguf`	Q8_0	Very High	Full 8-bit precision, closest to 16-bit original.

🚀 Usage

llama.cpp CLI

Make sure you are using a recent version of llama.cpp:

bash

llama-cli \
  -m Ornith-1.5-35B-A3B-abliterated-Q4_K_M.gguf \
  -p "You are a helpful assistant." \
  -c 32768 \
  -ngl 99 \
  --temp 0.7

LM Studio / Ollama / Other UI

Copy the repo link: NjProVk/Ornith-1.5-35B-A3B-abliterated-GGUF
Download your preferred quantization.
Configure context length and GPU layers to fit your hardware.

⚙️ Recommended Settings

Temperature: 0.6 – 0.8 (Lower for coding/logic, higher for creative roleplay/writing)
Top-P: 0.9 – 0.95
Min-P: 0.05

⚠️ Disclaimer

Uncensored Outputs: This model is fully uncensored and will not refuse sensitive prompts. It may generate controversial, adult, or unfiltered content.
Responsibility: The user assumes full responsibility for any content generated and must comply with applicable local laws.
Intended Use: For research, roleplay, creative writing, and testing in controlled environments.

NjProVk/Ornith-1.5-35B-A3B-abliterated-GGUF

Ornith-1.5-35B-A3B-abliterated-GGUF