Transformers Programming Tutorials & Engineering Articles

A central challenge in robotics is building policies that generalize beyond the demonstrations they’re trained on. A policy that succeeds in a training scene…

Hierarchical Interest Representation is a research area for Meta Ads. We’re exploring an upstream representation layer over the universe of Ads entities – users, advertisers, products, services – l…

What if autonomous coding AI agents could push your vision reasoning models above 90% accuracy with almost no manual effort? When adapting vision reasoning…

As context windows grow longer, moving large model weights efficiently becomes critical to performance. A common way to address this is quantization…

Generative AI workloads are rapidly outgrowing the memory and compute budget of single GPUs. For inference developers building media generation pipelines…

Every swipe, transfer, and payment on a modern financial network encodes a pattern of human behavior. Transaction data is one of the richest signals an…

Transformer architectures are the backbone of many modern large language and generative AI models. As these models grow in size, training runs consume more GPU…

Quick glossary for readers new to VLA/WAM terminology VLA Vision-Language-Action model: a robot policy that starts from a pretrained VLM backbone and adapts it…

Developers building real-time AI—such as chat assistants, copilots, and agentic workflows—are often constrained by token-by-token generation speed.

As enterprise AI adoption scales, developers are increasingly forced to stitch together fragmented pipelines—separate models for text, vision…

DiffusionGemma is an experimental text-generation model built on the Gemma 4 architecture that uses diffusion-based parallel generation instead of token-by-token autoregression, enabling much faster inference, bidirectional context awareness, and real-time self-correction while remaining deployable on consumer GPUs. Its architecture generates and refines 256-token blocks in parallel through iterative denoising, allowing it to handle complex constraint-based tasks such as Sudoku more effectively than traditional language models and demonstrating strong gains from fine-tuning. The model integrates with vLLM and other popular inference frameworks, giving developers access to a new non-autoregressive approach that combines high performance, efficient long-context scaling, and straightforward customization and deployment.

The newly released Gemma 4 12B is a dense, multimodal model designed for high-performance local AI execution on consumer devices. By introducing a novel, encoder-free architecture, it bypasses traditional visual and audio encoders to feed multimodal data directly into the LLM backbone.

Physical AI systems must understand the real world before they can act within it. Robots, autonomous vehicles, and smart spaces need to understand what’s…

Integration of Arm Scalable Matrix Extension 2 (SME2) and the Google AI Edge software stack enables high-performance, on-device generative AI by turning the CPU into a powerful matrix-compute accelerator. Using Stability AI’s "stable-audio-open-small" model as a case study, it outlines a streamlined "Convert, Optimize, and Deploy" pipeline that utilizes LiteRT, XNNPACK, and KleidiAI to automate hardware acceleration. The resulting implementation achieves over a 2x speedup in audio generation and a 4x reduction in memory usage while maintaining high audio quality on Arm-powered mobile devices and laptops.

For decades, computational biology has operated under a reductionist compromise. To fit complex biological systems into the limited memory of a single GPU…

In the current state of automotive radar, machine learning engineers can’t work with camera-equivalent raw RGB images. Instead, they work with the output of…

The article discusses NVIDIA TensorRT LLM AutoDeploy, a beta feature that automates the inference optimization process for large language models (LLMs).

The article discusses the limitations of current large language models (LLMs) in handling long contexts and introduces Test-Time Training with an end-to-end formulation (TTT-E2E) as a solution.

The article discusses the introduction of NVIDIA TensorRT Edge-LLM, an open-source C++ framework designed for high-performance inference of Large Language Models (LLMs) and Vision Language Models (...

This article provides a comprehensive tutorial on building a voice agent using NVIDIA's Nemotron models, focusing on retrieval-augmented generation (RAG) and safety guardrails.

The article discusses the evolution and scaling of Uber's Delivery Search Platform, emphasizing the transition from traditional lexical search to a semantic search model that enhances user experien...

The article discusses how to scale biology transformer models using PyTorch and NVIDIA BioNeMo Recipes, focusing on advanced parallel computing techniques and the integration of the NVIDIA Transfor...

The article discusses how to fine-tune the Gemma 3 270M model for on-device applications, enabling developers to create custom AI models without the need for expensive hardware.

The article provides an in-depth exploration of the EmbeddingGemma architecture, detailing its origins, embedding generation process, and the comprehensive training methodology.

The article discusses the challenges of cold start latency in deploying large language models (LLMs) and introduces the NVIDIA Run:ai Model Streamer, an open-source Python SDK designed to optimize ...

EmbeddingGemma is an innovative open embedding model designed for on-device AI applications, featuring 308 million parameters for efficient performance.

The article discusses fine-tuning the gpt-oss model for improved accuracy and performance through Quantization Aware Training (QAT) and Supervised Fine-Tuning (SFT).

This article discusses the development and implementation of forecasting models aimed at improving driver availability at airports, which are critical to Uber's ridesharing ecosystem.

The article introduces Gemma 3 270M, a compact AI model designed for hyper-efficient task-specific fine-tuning.

NVIDIA has optimized OpenAI's gpt-oss models for accelerated inference performance on the NVIDIA GB200 NVL72 system, achieving up to 1. 5 million tokens per second (TPS).

The article discusses the development of Jetflow, a framework designed by Cloudflare's Business Intelligence team to manage complex data ingestion tasks efficiently.

The article introduces Gemma 3n, a mobile-first architecture designed for on-device AI, highlighting its multimodal capabilities and architectural innovations.

LMArena, in collaboration with NVIDIA and Nebius, has developed the Prompt-to-Leaderboard (P2L) model to evaluate the performance of large language models (LLMs) across various tasks.

The article discusses the advancements in protein sequence alignment using MMseqs2-GPU and NVIDIA NIM, highlighting their significance in accelerating drug discovery and structural prediction in pr...

The article discusses NVIDIA's advancements in molecular AI modeling through the introduction of cuEquivariance and NIM microservices, which enhance the speed and efficiency of training and inferen...

The article discusses the advancements in large language models (LLMs) focusing on the importance of extended context lengths for processing and generating text.

The article discusses JUDE, LinkedIn's platform for generating high-quality embeddings for job recommendations using fine-tuned Large Language Models (LLMs).

The article discusses how to accelerate Deep Learning (DL) and Large Language Model (LLM) inference using Apache Spark in cloud environments.

The article discusses the new features and improvements in Gemma 3, highlighting its vision-language capabilities, architectural changes for memory efficiency, and enhanced multilingual support.

This article introduces the fundamental concepts of large language model (LLM) inference benchmarking, focusing on key metrics such as throughput and latency.

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 launch of ShieldGemma 2, a safety content classifier model built on Gemma 3, aimed at detecting harmful content in both synthetic and natural images.

The article discusses the advancements in AI-driven biological research with the introduction of Evo 2, a foundation model that integrates genomic, RNA, and protein data across multiple life domain...

PaliGemma 2 mix is an advanced vision-language model designed for multiple tasks, allowing developers to utilize a single model for various applications such as image captioning, object detection, ...

The article discusses advancements in embedding-based retrieval at Pinterest's Homefeed, focusing on improvements such as feature crossing, ID embeddings, and serving corpus upgrades.