Runtime format audit (2026-10-06)

No quantization-container correction was needed. This family has no n-gram tensors; no n-gram file or declaration was added. This remote format audit does not grant an oMLX/MTPLX runtime profile or a successful load/generation claim.

See runtime_audit.json for pinned config/index/header bindings, architecture, physical-format findings, and applied corrections. This is development evidence; no quality, MTP exactness, speed, or certification claim is added. Historical evidence stays bound to its original revision.

AX-Nemotron-3-Super-120B-A12B-MLX-AXQ-MXFP4-MTP

This repository contains an AXQuant development conversion of the pinned NVIDIA BF16 source. The main checkpoint uses standard MLX-LM config, tokenizer, index, and safetensors files. MXFP quantization is applied to eligible backbone weights; protected tensors keep their declared precision.

Source and format

• Source: nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16
• Immutable source revision: 2dc98e2afe4face0e4ce40972a915c45368bd34a
• AXQuant physical format: MXFP4
• AXQuant plan and file hashes are included in this repository.
• This is a development artifact; no quality certificate is claimed.

MTP payload

The source integrates Nemotron-H MTP tensors into its original checkpoint shards. AXQuant preserves those tensor payloads byte-for-byte in mtp.safetensors and records the source and payload digests in ax_nemotron_mtp_manifest.json. Runtime compatibility is unverified. This package does not claim MTP execution support in MLX-LM, AX Engine, MTPLX, or oMLX; each runtime owns that support.

Load the standard MLX backbone

from mlx_lm import load, generate
model, tokenizer = load("AutomatosX/AX-Nemotron-3-Super-120B-A12B-MLX-AXQ-MXFP4-MTP")
print(generate(model, tokenizer, prompt="Hello", max_tokens=32))

The example loads the text backbone. It does not load or activate the separate MTP payload. Super needs substantial memory; actual use depends on format, context, and runtime.

License and notices

The source uses the NVIDIA Nemotron Open Model License; included as NVIDIA-Nemotron-Open-Model-License.pdf.