NVIDIA released Nemotron-4 3B, a compact open-source LLM for edge and mobile deployment, trained on 8T tokens with reaso

NVIDIA released Nemotron-4 3B, a compact open-source LLM for edge and mobile deployment, trained on 8T tokens with reasoning benchmarks showing 72% on MATH and 68% on GSM8K. The model is available under Apache 2.0 via huggingface.co/nvidia.