NovaCorp releases Imp-RPG.System-1B, a 1B unrestricted Llama merge

Imp-RPG.System-1B is a 1-billion-parameter text-generation model from NovaCorp, built by merging Llama architecture weights using Mergekit and packaged in Safetensors format. The checkpoint landed on HuggingFace on July 3, 2026, tagged not-for-all-audiences to signal it lacks safety tuning and will generate unrestricted text when prompted. At 1 billion parameters, the model sits in the same weight class as Qwen2.5-1.5B-Instruct and TinyLlama-1.1B, small enough to run on consumer hardware with 4GB VRAM or less.

Mergekit has become the standard tool for combining pretrained checkpoints without retraining from scratch—practitioners blend base models, instruction-tuned variants, and domain-specific fine-tunes to produce hybrids optimized for niche tasks. Imp-RPG.System-1B targets role-play and creative-writing workflows that run outside content filters, following recent merges like Kunoichi-DPO-v2-7B and RP-Llama-3-8B. The HuggingFace card offers no benchmark numbers, training lineage, or recommended prompting strategy, a common pattern for early merge experiments where creators prioritize speed-to-release over documentation. The model is available under the Llama license terms inherited from its base weights.