Arsenic-Shahrazad 12B v4.3 GGUF quantization drops on HuggingFace
Quantized GGUF weights for the 12-billion-parameter Arsenic-Shahrazad v4.3 model are now available under a Creative Commons non-commercial license.
A new set of quantized weights has arrived for practitioners running uncensored language models locally: mradermacher's GGUF conversion of Arsenic-Shahrazad 12B v4.3 went live on HuggingFace this week.
Arsenic-Shahrazad 12B v4.3, a 12-billion-parameter language model from lambent, carries HuggingFace's "not-for-all-audiences" tag—the platform's marker for unrestricted or adult-capable models. The GGUF quantization makes the weights compatible with llama.cpp, Ollama, LM Studio, and other CPU/GPU inference engines that read the GGUF format. The model card lists the license as CC-BY-NC-4.0, which permits use and modification for non-commercial purposes with attribution.
GGUF quantizations compress full-precision weights into lower-bit representations—typically 4-bit, 5-bit, or 8-bit—to reduce memory footprint and speed up inference on consumer hardware. The mradermacher namespace on HuggingFace has become a go-to source for community GGUF conversions of open-weight models, particularly those in the uncensored or fine-tuned storytelling categories. Arsenic-Shahrazad's "not-for-all-audiences" tag signals that the model runs without built-in content filtering, a key feature for users who want full prompt control.