NVIDIA AI Releases Star Elastic: One Checkpoint that Contains 30B, 23B, and 12B Reasoning Models with Zero-Shot Slicing

The guide provided outlines the steps to install, load, and serve NVIDIA's Nemotron Labs 3 Elastic model in different scenarios. Here’s a summary of each step:

Step-by-Step Guide

Slide 1: Install Dependencies

Prerequisites

• Option A (Recommended for Production Serving): pip install vllm
• Option B (For Local Experimentation): pip install transformers torch accelerate and optionally, log in to Hugging Face if needed (pip install huggingface_hub huggingface-cli login).

Slide 2: Model Loading

Model Loading

• Load the single checkpoint containing all three nested variants (30B BF16, 23B BF16, and 12B BF16) using AutoTokenizer.from_pretrained() and AutoModelForCausalLM.from_pretrained().

python
 
1model_id = "nvidia/NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B-BF16"
2tokenizer = AutoTokenizer.from_pretrained(model_id, trust_remote_code=True)
3model = AutoModelForCausalLM.from_pretrained(model_id, trust_remote_code=True, torch_dtype
4
5[Read the full article at MarkTechPost](https://www.marktechpost.com/2026/05/09/nvidia-ai-releases-star-elastic-one-checkpoint-that-contains-30b-23b-and-12b-reasoning-models-with-zero-shot-slicing/)
6
7---
8
9**Want to create content about this topic?** [Use Nemati AI tools](https://nemati.ai) to generate articles, social posts, and more.