HydPy is hosting an NVIDIA Nemotron 3 Super Workshop - India

​Mark your calendars!

​HydPy is hosting an NVIDIA Nemotron 3 Super Workshop
This interactive workshop dives deep into Nemotron 3 Super, NVIDIA’s newly released 120B (12B active-parameter) open hybrid Mamba-Transformer Mixture-of-Experts (MoE) model. Designed to solve complex, dense technical problems autonomously, Nemotron 3 Super handles long-context analysis, precise reasoning, and coding while remaining computationally efficient. Throughout this course, participants will explore the model's architectural innovations such as Latent MoE, Multi-Token Prediction (MTP), and its hybrid backbone and learn how to customize, optimize, and deploy it using NVIDIA's fully open weights, datasets, and recipes.

​Full agenda details (TBA)
Agenda

• ​8:30 AM – 9: 30 AM
  Registration

• ​9: 30 AM - 9:45 AM
  HydPy Introduction

• ​9:45 AM - 10:15 AM
  NVIDIA Nemotron Strategy: Open Datasets, Weights, Frameworks and Recipes.

  • ​Speaker : Megh Makwana

    • ​Manager, Applied Gen AI Solution Engineering, NVIDIA

    • ​Megh’s work converges on building foundational AI models, scaling GPU workloads efficiently, and supporting CSPs in developing AI platforms with NVIDIA AI Enterprise.

• ​10:15 AM - 10:45 AM
  Deep-Dive into Nemotron 3 Super Architecture Design

• ​10:45 AM - 11:30 AM
  Efficient Serving & Deployment of Nemotron Model using vLLM, SGLang and TensorRT-LLM Cookbooks

• ​11:30 AM - 11:45 AM
  Break

• ​11:45 AM – 01:00 PM
  Building Agentic Workflow using Nemotron models

• ​1:00 PM - 1 :45 PM
  LUNCH

• ​1:45 PM - 2 :15 PM
  Use case/Demo by Customer/Partner

• ​2:15 PM - 3:00 PM
  Talk from HydPy

• ​3 PM - 4 :00 PM
  Paper Presentation on NVIDIA Nemotron 3 Super

• ​4:00 PM - 4:15 PM
  Snack Break

• ​4:15 PM - 4:30 PM
  Project & Blog Contest Announcement by HydPy

• ​4:30 PM Onwards
  Engagement Activities with HydPy,Winners Announcements, Open Discussion , Networking

​Prerequisites:

• ​Languages/tools—Python

• ​Frameworks— PyTorch, TensorRT-LLM, Triton Inference Server, SGLang, vLLM.

• ​Sign-up with build.nvidia.com and create your account

• ​Bring your Laptop to this Workshop
  Laptop with internet access - Ideal minimum: 5 Mbps download / 1–2 Mbps upload. This will ensure consistent access to the lab. This will ensure consistent access to the lab.

​Showcase Your NVIDIA Nemotron 3 Super ?
Are you interested in presenting your work, project, or paper related to NVIDIA Nemotron 3 Super?
Submit your details by 1st May. Selected participants will get an opportunity to present, and the best presentation will be rewarded on the event day.
Apply here: https://forms.gle/7AGSHWr4XiC66sxo7

​Note: Filling this form you will provide consent to share your details with the host company, IIIT-H and NVIDIA for the purpose of event coordination and communication. You will receive the Invite from LUMA 1-2 days before the event that will facilitate check-in process.

​Venue:
International Institute of Information Technology, Hyderabad
Professor C. R. Rao Road, Gachibowli, Hyderabad – 50003

​Want to know more about NVIDIA Nemotron 3 Super?

​Explore the following resources to understand the architecture, capabilities, and real-world applications of NVIDIA Nemotron 3 Super:

​Want to know more about NVIDIA Nemotron 3 Super?

​Explore the following resources to understand the architecture, capabilities, and real-world applications of NVIDIA Nemotron models:

​NVIDIA Nemotron NVIDIA Nemotron Models

​NVIDIA Nemotron 3 Super Model Blog Introducing Nemotron 3 Super: An Open Hybrid Mamba-Transformer MoE for Agentic Reasoning

​NVIDIA Nemotron 3 - GitHub Repo NVIDIA-Nemotron-3-Super-120B-A12B-FP8

​NVIDIA Nemotron 3 - Tech Blog New NVIDIA Nemotron 3 Super Delivers 5x Higher Throughput for Agentic AI

​NVIDIA Nemotron 3 Tutorial Nemotron 3 Super Tutorial: Multi-Token Prediction, Latent MoE, Perplexity and OpenCode Integration

​Hugging Face Get started with Nemotron 3 Nano on Hugging Face

​Target Audience
This workshop is crafted for:
AI/ML Engineers & Researchers: Professionals building advanced LLMs and specialized reasoning models.

​Generative AI Developers: Developers creating complex, multi-agent AI applications (such as software development or cybersecurity triaging agents).

​Data Scientists: Practitioners interested in state-of-the-art training techniques like Native NVFP4 pretraining and multi-environment reinforcement learning (RL).

​DevOps & MLOps Engineers: Professionals looking to deploy high-throughput, low-latency AI models using vLLM, SGLang, or NVIDIA TensorRT-LLM.

​Participants should have basic knowledge of Python, containerized environments, and experience working in Jupyter/Colab or similar notebook workflows.

​Why Should Someone Attend It?

• ​Accelerate / Optimize LLM Inference: Learn how to maximize compute efficiency without sacrificing accuracy. You will explore how Nemotron-3 Super achieves a 5x throughput increase over previous generations. We will dive into architectural breakthroughs like Latent MoE (which calls 4x as many experts for the exact same inference cost), Multi-Token Prediction (MTP) for built-in speculative decoding and 3x faster generation, and memory-saving Native NVFP4 pretraining.

• ​Industry Relevant Use-Cases: Move beyond simple chatbots and into the realm of autonomous, multi-step AI agents. You will discover how to apply the model's massive 1M-token context window to real-world scenarios like dense software development and cybersecurity triaging. We will also cover the highly efficient "Super + Nano" deployment pattern, teaching you how to intelligently route simple tasks to smaller models and complex planning to the 120B Super model to optimize cloud costs.

• ​End-to-End Practical Skills: Walk away with the hands-on experience needed to build and deploy immediately. Because Nemotron 3 Super is fully open, you will learn how to leverage NVIDIA’s open weights, datasets, and recipes. You will get practical experience using deployment cookbooks (vLLM, SGLang, TensorRT-LLM) for low-latency production, and explore fine-tuning techniques (LoRA/SFT, GRPO/DAPO) using the NeMo ecosystem to customize the model for your specific proprietary data.

​Learning Objectives :
By the end of the workshop, attendees will be able to:

• ​Understand Advanced Architectures: Grasp the mechanics behind the Hybrid Mamba-Transformer backbone, Latent MoE (which calls 4x as many experts for the same compute cost), and Multi-Token Prediction (MTP) for built-in speculative decoding.

• ​Navigate the Training Pipeline: Understand how the model achieves stability and accuracy through Native NVFP4 pretraining and trajectory-based reinforcement learning across diverse environments (using NeMo Gym and NeMo RL).

• ​Implement the "Super + Nano" Pattern: Learn how to architect multi-agent workflows that smartly route simple tasks to Nemotron 3 Nano and complex planning/reasoning tasks to Nemotron 3 Super.

• ​Deploy and Fine-Tune: Utilize NVIDIA’s open resources and deployment cookbooks (vLLM, SGLang, TensorRT-LLM) to customize and run the model on your own infrastructure.

​About OSDG, IIIT Hyderabad

​OSDG (Open Source Development Group) is the leading technical club of IIIT Hyderabad, focused on community building and open-source initiatives within one of India’s premier technical and research institutes. OSDG fosters a strong culture of collaboration, learning, and open contribution through hands-on projects, workshops, and community-driven programs. We are grateful to OSDG for their invaluable organizational and technical support in making this meetup possible. Learn more @ osdg.in

​Interested in Speaking at a Future HydPy Meetup?
Submit your proposal by creating a GitHub issue here: http://bit.ly/hydpy-cfp

​Join the Conversation
Website: hydpy.org
Twitter: @hydPython
Meetup: bit.ly/hydpy-meetup
Telegram: https://t.me/HydPy
LinkedIn: https://www.linkedin.com/company/hydpy/