IDEAHQ/ava-nautilus

AVA Nautilus Models (Nemotron)

On-device LLM models for VoiceOS, based on NVIDIA Nemotron.

Models

AVA ID	Source	Active Params	Format	Target
AVA-NAUTILUS-4B	Nemotron-3-Nano-4B	4B	safetensors + GGUF	Phone
AVA-NAUTILUS-9B	Nemotron-Nano-9B-v2	9B	safetensors	Desktop / tablet
AVA-NAUTILUS-30B-A3B	Nemotron-3-Nano-30B	3B active (MoE)	safetensors + GGUF	Phone (flagship)
AVA-NAUTILUS-120B-A12B	Nemotron-3-Super-120B	12B active (MoE)	safetensors	Desktop
AVA-NAUTILUS-CASCADE	Nemotron-Cascade-2-30B-A3B	3B active	safetensors + GGUF	Phone (reasoning)
AVA-NAUTILUS-VL-8B	Nemotron-Nano-VL-8B	8B	safetensors	Vision tasks

Directory Structure

raw/                           # Base model weights (safetensors)
  nemotron-3-nano-4b/
  nemotron-nano-9b-v2/
  nemotron-3-nano-30b-a3b/
  nemotron-3-super-120b-a12b/
  nemotron-cascade-2-30b-a3b/
  nemotron-nano-vl-8b/
gguf/                          # Quantized for on-device inference
  nemotron-3-nano-4b-Q4_K_M.gguf
  nemotron-3-nano-30b-A3B-Q4_K_M.gguf
  nemotron-cascade-2-30b-a3b-Q4_K_M.gguf
production/                    # AON-encrypted (deployed to devices)

License

Raw models: NVIDIA Open Model License. AON-encrypted production variants: Proprietary (Intelligent Devices LLC).