Nvidia releases a new small, open model Nemotron-Nano-9B-v2 with toggle on/off reasoning
1 min read
Summary
Following the release of small AI vision and language models from the likes of Liquid AI, Google and MIT, Nvidia has launched its own small language model (SLM), Nemotron-Nano-9B-V2, which claims to have the highest performance in its class on selected benchmarks.
The 9bn parameter model runs on a single Nvidia A10 GPU and can handle multiple languages, including English, German, Spanish and French, as well as having the ability for users to toggle on and off AI reasoning, or self-checking before outputting an answer.
The model is based on Nemotron-H, a set of hybrid Mamba-Transformer models and can handle very long sequences of information in and out by maintaining state, making it more efficient.
It boasts competitive accuracy against other open small-scale models and is suitable for both instruction following and code generation.