LocalLLaMA steers tablet users to tiny NSFW models

This is less a model recommendation than a hardware reality check: 8GB RAM on a tablet can run something, but not the kind of uncensored roleplay experience most newcomers imagine. The thread's concrete suggestions are Satyr-V0.1-4B, Impish_LLAMA_4B, BlackSheep-Llama3.2-3B, and Gemmasutra-Mini-2B-v1, which back up the tiny-model framing. Satyr is explicitly tagged uncensored/NSFW, Impish ships GGUF plus ARM-friendly mobile quantizations, Gemmasutra says it works on browser, laptop, phone, and Raspberry Pi, and one reply sets the real floor at 16GB+ RAM with a Mistral Nemo-based model if the goal is decent RP instead of a proof of concept. The cleanest user experience is likely remote inference: run the model on a PC or server, then use the tablet only as the client, because heat, battery drain, and latency will bite long before the model feels smart.