NVIDIA Releases Nemotron 3.5 Super VL 120B Model
NVIDIA launches Nemotron 3.5 Super VL, a 120B open multimodal MoE model featuring a 1-million-token context window and NVFP4 weights.
Stock photo for illustration only, not from the actual event
- NVIDIA releases the open Nemotron 3.5 Super VL model with 120B parameters
- Features a multimodal MoE architecture with 12B active parameters
- Supports a massive 1-million-token context window using NVFP4 weights
- Runs on a single DGX Spark alongside two additional edge models
NVIDIA has officially announced the release of its newest artificial intelligence model, Nemotron 3.5 Super VL. This open multimodal Mixture-of-Experts (MoE) model comes with a total parameter count of 120 billion, engineered to handle diverse data types with advanced multimodal capabilities.
A key highlight of this release is its efficiency, utilizing only 12 billion active parameters during inference while maintaining high performance. Furthermore, the model supports an extensive context window of 1 million tokens, allowing it to process and analyze vast amounts of data seamlessly.
Stock photo for illustration only, not from the actual event
Regarding deployment, the model utilizes NVFP4 weights and is optimized to run on just one DGX Spark system. In addition to the primary model, NVIDIA has introduced two separate edge models to broaden accessibility and performance for resource-constrained environments.
The introduction of Nemotron 3.5 Super VL highlights NVIDIA's ongoing push to make large-scale AI accessible to developers via open-source releases. By combining MoE architecture with NVFP4 weight optimization, organizations can operate a 100B-plus parameter model without requiring massive server clusters, significantly lowering infrastructure barriers.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment