Nemotron-Mini-4B

Nemotron-Mini-4B-Instruct is a small language model NVIDIA built by pruning and distilling its 15B Nemotron-4 model down to 4 billion parameters, then instruction-tuning it for roleplay, retrieval-augmented question answering, and function calling. It runs in a 4,096-token context window and was optimized for speed and on-device deployment.

The model ships under the NVIDIA Community Model License, whose model card describes it as ready for commercial use, though teams should verify the exact terms before shipping products. It has become a popular base for fine-tuning experiments and local chatbot projects because it fits comfortably on consumer GPUs and Apple Silicon Macs.