NVIDIANemotron 3.5
Nemotron 3.5 Lightning 30B-A3B
Lightning combines Mamba-2, expert and attention layers. It has 30B total parameters and 3B active parameters.
What the release provides
The BF16 card supports context up to one million tokens. Its single-H100 recipe uses 256K instead. The release includes separate speculative-decoding approaches and uses OpenMDW 1.1.Nemotron 3.5 Lightning 30B-A3B official model card
How to use it well
Practical guidance. Use the serving recipe that matches your hardware and concurrency. The largest advertised context is not the capacity of every deployment. Measure latency on several simultaneous conversations before choosing a decoding method.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16 |
| Release license | OpenMDW 1.1Nemotron 3.5 Lightning 30B-A3B official model card |