Kapitsa Lightning

Nemotron 3.5 Lightning 30B-A3B (NVFP4, 22.9 GB) running locally on Apple Silicon through Kapitsa. Fast replies on 32 GB Macs and up.

kapitsa serve --model lightning

Speed

Mac Prefill (tok/s, 4K prompt) Decode (tok/s)
M5 Max 128 GB 5,738 287
M5 Pro 48 GB 3,141 159
M4 Pro 48 GB 945 112
M3 Pro 36 GB 788 75

Architecture

Hybrid Mamba-2 MoE, 30B total parameters, 3B active per token, NVFP4 quantization.

Weights

nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-NVFP4

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support

Model tree for nace-ai/kapitsa-lightning

Finetuned
(9)
this model