Nvidia · Nemotron-Flash
Nemotron-Flash-1B
Nvidia's 0.96B hybrid DeltaNet, Mamba2 and attention model, optimized around measured data-centre GPU latency.
- Parameters
- 960M
- Input
- Text
- Output
- Text
- Access
- Non-commercial
- Licence
- CC BY-NC 4.0
- Released
- Device evidence
- No device evidence
- Last verified
- Formats
- Safetensors
- Runtimes
- Transformers, TensorRT-LLM
- Platforms
- Linux
Nvidia’s published latency and throughput results use H100 GPUs. The paper supplies no phone or embedded-device measurements, and the model needs implementations for its hybrid operators.