Arsenic-Shahrazad 12B v4.1 GGUF quantizations now available on HuggingFace
mradermacher published GGUF quantizations of lambent's Arsenic-Shahrazad 12B v4.1, an uncensored 12-billion-parameter model under CC-BY-NC-4.0.
Arsenic-Shahrazad 12B v4.1, a 12-billion-parameter uncensored language model from lambent, is now available in GGUF quantized format on HuggingFace under CC-BY-NC-4.0 license. The license restricts commercial use while permitting research and personal use. GGUF quantizations let practitioners run the weights on consumer hardware — a MacBook with 16GB RAM can typically handle Q4_K_M quants at conversational speed, and a desktop with 24GB VRAM can load Q5 or Q6 variants for higher fidelity.
The model carries a "not-for-all-audiences" tag, signaling that safety guardrails have been removed or never applied. At 12 billion parameters, the model sits in a practical sweet spot: large enough for coherent multi-turn conversations and complex instruction-following, yet small enough to run on a single consumer GPU or recent Apple Silicon laptop. Quantization compresses weights into 4-bit, 5-bit, or 6-bit representations, cutting memory requirements by 75 percent or more with minimal quality loss. mradermacher's repo includes Q4_K_M for maximum speed, Q6_K for near-lossless quality, and several intermediate options. The upload went live on May 16, 2025.