AWS Launches Serverless Fine-Tuning for NVIDIA Nemotron 3
AWS ML Blog · rss · 2026-07-10
Amazon SageMaker AI announced support for serverless customized fine-tuning of the NVIDIA Nemotron 3 model family, initially supporting the Nano (30B) and Super (120B) versions.
Model Features
Nemotron 3 uses a hybrid Mamba-Transformer MoE architecture and supports a context length of up to 1 million tokens. By activating only a small fraction of parameters (e.g., 12B for the Super version), it achieves high throughput and low computing costs, making it perfect for multi-agent workflows.
Fine-Tuning Techniques
Users can adapt the model to specific domains using three techniques without managing underlying GPU infrastructure:
- Supervised Fine-Tuning (SFT): Teaches new behaviors using high-quality input-output pairs.
- Reinforcement Learning with Verifiable Rewards (RLVR): Optimizes the model against verifiable targets like tool-calling accuracy and code correctness.
- Reinforcement Learning from AI Feedback (RLAIF): Uses AI models to evaluate outputs, aligning brand tone and improving quality on open-ended tasks.
More from Models
- Google launches three new Gemini models, including a cybersecurity system — Polymarket · 2026-07-22
- Google says information agents are coming to AI Pro and Ultra this summer — gaganghotra_ · 2026-07-22
- Poolside’s Laguna S 2.1 gets a two-week free run on Nous Portal — NousResearch · 2026-07-22
- Qwen3.8 Max Preview looks substantially better in a side-by-side test with Kimi K3 — curiousily_ · 2026-07-22
- Moonshot’s Kimi K3 reaches #5 on MathArena as the top open model — xeophon · 2026-07-22
- Google launches Gemini 3.5 Flash Cyber for CodeMender, with limited access for governments — GoogleAI · 2026-07-22