Downloads · 30 days
135
5% of all-time downloads
baa-ai/Nemotron-3-Super-120B-A12B-MINT-MLX
Nemotron-3-Super-120B-A12B-MINT-MLX is a machine learning model from baa-ai. Use it for the machine learning task on the model card, and read the license before you ship it in a product. It is set up for mlx. The card lists the license as other.
Mixed-precision quantized version of nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 optimised by baa.ai.
Downloads · 30 days
135
5% of all-time downloads
All-time downloads
2.9K
Public
Parameters
121B
105 GB on disk
Likes
1
Public
Click a slice to open those files.
.safetensors105 GB · 100%
How the weights are stored.
U32109B · 90%
From the Hugging Face model README
Mixed-precision quantized version of nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 optimised by baa.ai.
Hybrid Mamba-MoE-Attention architecture (512 experts, 22 active per token) quantized to fit a single Apple Silicon machine.
| Metric | Value |
|--------|-------|
| Size | 98 GB |
| Average bits | 7.0 |
| Framework | MLX |
| Architecture | Hybrid Mamba-2 + MoE + Attention |
| Parameters | 123.6B (12B active) |
| Precision | Parameters | Percentage |
|-----------|-----------|------------|
| 16-bit | 18.2B | 14.8% |
| 8-bit | 43.9B | 35.5% |
| 4-bit | 61.4B | 49.7% |
from mlx_lm import load, generate
model, tokenizer = load("baa-ai/Nemotron-3-Super-120B-A12B-MINT-MLX")
messages = [{"role": "user", "content": "Hello!"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
response = generate(model, tokenizer, prompt=prompt, max_tokens=256)
print(response)
Requires mlx-lm ≥ 0.31.1 with latent MoE projection support
Peak memory: ~100 GB (fits on M2 Ultra 192 GB or M4 Max 128 GB)
MTP (Multi-Token Prediction) layers are stripped for inference — they are not needed for standard autoregressive generation
Quantized by baa.ai
Black Sheep AI builds private AI with the Bounded AI Architecture (BAA): three layers, each bound to something you control, on infrastructure you control. Nothing leaves your boundary.
Ram (Reasoning Layer) — Full-capability frontier models, shrunk to fit smaller GPUs with reasoning kept and certified. They rely strictly on your Data Modules for knowledge, so you swap the model as better ones ship.
Paddock (Knowledge Plane) — Binds your Data Modules directly to the reasoning layer: one per topic, period, person, permission or tenant. Permission, freeze, hand back or delete each one on its own.
Watchman (GRC Layer) — Governance, risk and compliance in a single, complete audit log: every module permission, every bind, every model's provenance. Five certificates before anything goes live.
Shepherd — Takes a frontier model, compresses it into a Ram reasoning model, certifies it with Watchman, binds it to your Paddock, and deploys the Bounded AI Architecture to your infrastructure: your cloud, a Mac Studio, or an air-gapped facility.
Learn more at baa.ai — Private AI for enterprises whose data cannot leave the building.