Downloads · 30 days
0
Nemotron-3-Super-120B-A12B-ElasticBit-POC
maxzameer/Nemotron-3-Super-120B-A12B-ElasticBit-POC
Nemotron-3-Super-120B-A12B-ElasticBit-POC is a machine learning model from maxzameer. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as other.
This repository currently contains a layer-level research proof of concept, not a complete runnable Nemotron checkpoint.
Access
Public
Updated Sep 18, 2026
Repo size
2.3 GB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.elasticbit2.3 GB · 100%
At a glance
- License
- other
- Access
- Public
- Created
- Sep 18, 2026
- Updated
- Sep 18, 2026
- SHA
- 0ec2ebd6
Model card
From the Hugging Face model README
Nemotron 3 Super 120B-A12B — ElasticBit layer POC
This repository currently contains a layer-level research proof of concept, not a complete runnable Nemotron checkpoint.
Source
- Model:
nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 - Source precision: BF16
- ElasticBit working/reference precision: FP16-cast
- Target layer: 1 (first MoE layer)
- ElasticBit threshold: 0.01
- MLBricks Git commit:
f772eb27d2d6b84764e9e90458827dac6980a88a
Calibration safeguards
- Official NVIDIA Nemotron pretraining sample data was used.
- Calibration target inputs were generated through the real embedding and preceding Nemotron layer(s).
- Real non-zero routed-token coverage was required for all 512/512 routed experts.
- Inactive-expert synthetic zero calls were excluded from ElasticBit's calibration-row collection.
- Whole-layer output error was measured against the uncompressed FP16-cast reference layer.
See poc_report.json for the exact bit distribution, errors, cast error,
storage reduction and environment information.
The MLBricks/ElasticBit code is licensed under its repository license. The source model remains subject to NVIDIA's Nemotron Open Model License.
- License
- other
- Created
- Sep 18, 2026
- Updated
- Sep 18, 2026