2 min read
Showcasing LLM Performance: How Research Papers Present Evaluation Results
Showcasing LLM Performance: How Research Papers Present Evaluation Results
All Posts
Agentic AI
Agentic AI & LLM Engineering
Deep dives into agentic AI, LLM evaluation, synthetic data generation, model selection, and production AI engineering best practices.
5 min read
Why LLM Accuracy Is Won or Lost Before Training Begins: The Case for Data Curation
Data curation is the single biggest factor in LLM performance. Learn how NeMo Curator uses GPU-accelerated deduplication, synthetic data, and classification at scale.
5 min read
Why Synthetic Data Generation Is Critical for LLM Training in 2026
Synthetic data generation has become essential for training high-quality LLMs. Learn the generate-critique-filter pipeline that transforms raw data into production-grade training sets.
6 min read
The 6-Step Synthetic Data Pipeline for LLM Fine-Tuning and Alignment
Build a production-grade synthetic data pipeline for LLM fine-tuning and alignment with prompt critique loops, reward models, safety filtering, and practical examples.
6 min read
Synthetic Data Generation for RAG and Agentic AI: A Production Pipeline Guide
How to build a reliable synthetic data pipeline for RAG and agentic AI systems using the generate-critique-filter-curate workflow trusted by production AI teams.
6 min read
Inside the NeMo Curator Workflow: From Raw Web Text to Training-Ready LLM Data
A step-by-step breakdown of the NeMo Curator data curation pipeline for LLM pre-training — covering web crawling, deduplication, quality filtering, and decontamination.
6 min read
Document-Level Deduplication for LLM Training: Exact, Fuzzy, and Semantic Methods Explained
Master the three approaches to document-level deduplication — exact hashing, MinHash with LSH, and semantic embeddings — to improve LLM training data quality.
5 min read
What Is the Best Data Format for Fine-Tuning LLMs? A Complete JSONL Guide
JSONL is the standard data format for LLM fine-tuning. Learn why JSON Lines works best, how NeMo Curator processes raw data into JSONL, and best practices for training datasets.
6 min read
How to Create Synthetic Data for LLM Training with NeMo Curator: Pipelines and APIs
NeMo Curator provides GPU-accelerated synthetic data generation pipelines for LLM training. Learn the Open QA, Writing, Math, and Coding pipelines with practical examples.
6 min read
NeMo Curator Classifier Models: How Domain and Quality Classification Creates High-Quality Data Blends
NeMo Curator's Domain Classifier and Quality Classifier use GPU-accelerated RAPIDS to split LLM training data into balanced, high-quality blends at terabyte scale.
6 min read
Why Data Curation for LLM Training Takes So Long: Text, Image, and Video Processing Bottlenecks
Traditional data curation pipelines for LLM training face critical bottlenecks in synthetic data generation, quality filtering, and semantic deduplication across text, image, and video modalities.
5 min read
Quality Data Filtering vs Fuzzy Deduplication: The Critical Tradeoff in LLM Training
Learn how quality filtering and fuzzy deduplication create a tradeoff in LLM data curation, and how NeMo Curator uses GPU acceleration to handle both at scale.
4 min read
How NVIDIA NeMo Curator Speeds Up LLM Training: Benchmarks and Results
NeMo Curator delivers 17x faster data processing with measurable accuracy gains. See the GPU scaling benchmarks and real-world performance improvements for LLM training.
5 min read
Azure AI Foundry Agent Service: A Complete Guide to Building Enterprise AI Agents
Azure AI Foundry Agent Service provides a managed framework for building, managing, and deploying AI agents on Azure. Compare it to Semantic Kernel, AutoGen, and Copilot Studio.
5 min read
AI Agents: What They Are and the Current Landscape in 2025
A comprehensive overview of AI agents — what they are, how they work, and the major platforms including GPT Agents, Gemini, Claude, Copilot, AutoGen, and AutoGPT.
6 min read
Prompt Task Classification and Complexity Evaluation: NVIDIA's DeBERTa-Based Framework Explained
NVIDIA's prompt-task-and-complexity-classifier categorizes prompts across 11 task types and 6 complexity dimensions using DeBERTa. Learn how it works and when to use it.
5 min read
Retrieval-Augmented Generation (RAG): How It Works and Why It Matters
RAG strengthens LLM responses by grounding them in external knowledge sources. Learn how retrieval-augmented generation reduces hallucinations and enables real-time knowledge access.