Kapitsa Dory
Nemotron Dory 4B (NVFP4, 3.2 GB) running locally on Apple Silicon through Kapitsa. A looped dense model that fits on any Mac.
kapitsa serve --model dory
Speed
| Mac | Prefill (tok/s) | Decode (tok/s) |
|---|---|---|
| M5 Max 128 GB | 122 | |
| M5 Pro 48 GB | 70 | |
| M4 Pro 48 GB | 64 | |
| M3 Pro 36 GB | 450 | 36 |
| M1 Pro 16 GB | 39 |
Architecture
Looped dense transformer, 4B parameters, NVFP4 quantization.
Weights
Model tree for nace-ai/kapitsa-dory
Base model
nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-NVFP4