Reinforcement Learning Programming Tutorials & Engineering Articles

How we compress production failures into model weights every day, beat frontier-model quality, and cut serving costs 96%.

Customization is what enables developers to take a general model and tailor it to use cases, domains, languages, and more. However, customization comes with a…

Tunix is Google’s new JAX-native post-training library designed to eliminate TPU idling bottlenecks when training multi-turn, tool-using LLM reasoning agents. It maximizes hardware throughput by combining highly concurrent, asynchronous rollouts with a decoupled producer-consumer pipeline, ensuring the trainer is constantly fed even while agents wait on network I/O or environment steps. Additionally, Tunix provides plug-and-play abstractions and continuous macro-level profiling, allowing developers to easily integrate custom open-source environments and optimize complex distributed workflows without massive code rewrites.

Robotics foundation models have made remarkable progress. Today’s best systems can follow natural language instructions to pick, place, sort…

Reinforcement learning (RL) is central to aligning language models, from reinforcement learning with human feedback (RLHF) within AI assistants to newer…

Single-turn chatbots are evolving into long-running agents that can reason, maintain context, use tools, and run efficiently across many turns to complete…

The Google Tunix Hackathon on Kaggle challenged developers to transform small, non-reasoning base models into general reasoning engines using Kaggle TPUs and a limited compute budget. The winning teams achieved this by implementing multi-stage post-training pipelines that combined Supervised Fine-Tuning (SFT) with advanced alignment techniques like GRPO and SimPO. Ultimately, the competition democratized AI development by proving that highly capable, structured reasoning models can be successfully trained by the community using accessible, open-source resources.

As LLMs transition from simple text generation to complex reasoning, reinforcement learning (RL) plays a central role. Algorithms like Group Relative Policy…

MaxText has introduced new support for Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL) on single-host TPU configurations, leveraging JAX and the Tunix library for high-performance model refinement. These features enable developers to easily adapt pre-trained models for specialized tasks and complex reasoning using efficient algorithms like GRPO and GSPO. This update streamlines the post-training workflow, offering a scalable path from single-host setups to larger multi-host configurations.

This is the second post in the Ranking Engineer Agent blog series exploring the autonomous AI capabilities accelerating Meta’s Ads Ranking innovation. The previous post introduced Ranking Eng…

This article explores how to train an AI agent to operate a new Command Line Interface (CLI) using synthetic data generation and reinforcement learning.

The article discusses the NVIDIA Nemotron 3, a family of open models designed for agentic AI systems, emphasizing its efficiency and accuracy through innovative architectures and techniques.

The article discusses the development of scientific AI agents using reinforcement learning (RL) techniques, specifically through the NVIDIA NeMo framework.

The article introduces Broadened Reinforcement Learning (BroRL), a new paradigm that enhances the training of large language models (LLMs) by focusing on rollout scaling rather than just increasing...

Netflix introduces Advantage-Weighted Supervised Fine-Tuning (A-SFT), a novel post-training algorithm for generative recommender systems that addresses the unique challenges of applying reinforceme...

The article introduces Tunix, a new open-source, JAX-native library designed for post-training of large language models (LLMs).

This article discusses the integration of the Newton physics engine with NVIDIA Isaac Lab for training quadruped locomotion policies and simulating cloth manipulation.

The article discusses the enhancements in reinforcement learning training throughput using NVIDIA NeMo-RL with Megatron-Core support.

The article discusses the advancements in reinforcement learning for large language models (LLMs) through the introduction of ProRL v2 by NVIDIA Research.

The article discusses the release of the NVIDIA Llama Nemotron Super 49B v1. 5, highlighting its advancements in accuracy, efficiency, and reasoning capabilities for AI agents.

The article introduces NVIDIA NeMo-RL, an open-source library for reinforcement learning that supports scalable training from single-GPU to thousand-GPU models.

This article discusses how Uber utilizes reinforcement learning techniques to enhance the efficiency of its marketplace by improving the balance between drivers and demand.

The article discusses the advancements in AI autonomy through NVIDIA's Nemotron open reasoning models, which enhance AI agents' decision-making capabilities in complex environments.

The article discusses the development and capabilities of NVIDIA's Llama Nemotron reasoning models, which enhance AI agents' reasoning abilities for complex problem-solving in various industries.

The article discusses how NVIDIA Cosmos World Foundation Models (WFMs) enhance the development of AI-driven robots and autonomous vehicles by providing high-fidelity, physics-aware synthetic data.

Gemma 3 is the latest version of the Gemma open-model family, boasting enhanced capabilities such as multimodality, longer context windows, and improved reasoning.

The article discusses the development of domain-adapted foundation GenAI models at LinkedIn, focusing on their application within the Economic Opportunity Network (EON) project.

The article discusses Airbnb's evolution in location retrieval from simple heuristics to advanced machine learning and reinforcement learning techniques.

The article discusses Netflix's approach to enhancing long-term member satisfaction through its recommendation algorithms.

The article discusses the development of Pinterest Canvas, a text-to-image foundation model aimed at enhancing existing images and products on the Pinterest platform.

The article discusses the challenges and methodologies involved in training quadruped locomotion policies using NVIDIA Isaac Lab, emphasizing the importance of high-fidelity simulation for bridging...

The article discusses the advancements in robotics workflows through the latest release of NVIDIA Isaac Sim 4. 0 and NVIDIA Isaac Lab.

The article discusses the integration of AI in enabling practical quantum computing by addressing challenges in quantum processors, error correction, and algorithm development.

The article introduces DRaFT+, an enhanced algorithm for fine-tuning text-to-image diffusion models, which aims to improve the alignment between input prompts and generated images.

The article discusses Yoko Li's innovative work in AI, focusing on her projects like AI Town and AI Tamago, which utilize emergent behavior and large language models.

The article discusses how to leverage Palantir AIP to build a semantic search application that uncovers insights from unstructured data within enterprises.

The article discusses how NVIDIA NeMo can streamline the development of generative AI applications on GPU-accelerated Google Cloud.

QHack 2023 showcased the intersection of quantum computing and machine learning, featuring 2,850 participants from 105 countries competing to develop innovative solutions using NVIDIA's quantum tec...

The article discusses how AutoDMP leverages AI and GPU technology to optimize macro placement in chip design, significantly improving performance and efficiency.

The article summarizes the advancements and AI-powered solutions introduced in 2022, highlighting the most popular posts on the NVIDIA Technical Blog.

The article discusses the DeXtreme project, which utilizes simulation to teach dexterity to a real robot hand.

This article discusses the application of reinforcement learning to create optimal recommendation systems that consider users' time budgets.

NVIDIA researchers introduced Factory, a novel simulation approach designed to enhance robotic assembly by enabling real-time, accurate simulations of contact-rich interactions.

The article discusses the innovative use of deep reinforcement learning (RL) to design arithmetic circuits, particularly in the context of NVIDIA GPUs.

The article discusses NVIDIA's advancements in MLPerf Training v2. 0, highlighting the full-stack optimizations that enhance performance across various AI workloads.

The article discusses the performance improvements achieved in the NVIDIA MLPerf Training v1. 1 benchmark through full stack optimization.

NVIDIA has released an updated Edge AI and Robotics Teaching Kit aimed at university educators, developed in collaboration with experts from the University of Oxford and the University of Maryland,...

The article discusses NVIDIA's research on transferring dexterous manipulation capabilities from GPU simulation to real-world robotic applications.

This article discusses how Airbnb utilizes task-oriented conversational AI to enhance customer support for hosts and guests.