NVIDIA Certified Associate - Generative AI LLMs
Validates foundational knowledge of generative AI and large language model development using NVIDIA platforms, including core machine learning concepts, neural network architectures, transformer models, LLM fine-tuning with NeMo, inference optimization with TensorRT-LLM, model serving with Triton, RAG architectures, prompt engineering, experiment design, data preprocessing, and trustworthy AI with NeMo Guardrails. The exam covers five domains: Core Machine Learning and AI Knowledge (30%), Software Development (24%), Experimentation (22%), Data Analysis and Visualization (14%), and Trustworthy AI (10%). Format: 50-60 multiple-choice questions, 60 minutes, proctored online.
Exam domains
- Core Machine Learning and AI Knowledge30%
Foundational LLM and generative AI concepts including transformer architectures, tokenization, embeddings, prompt engineering, and retrieval-augmented generation. Covers full supervised fine-tuning vs PEFT/LoRA trade-offs in NeMo AutoModel and alignment techniques such as SteerLM, RLHF, and DPO.
- Software Development24%
Building and deploying generative AI applications on NVIDIA NeMo, NIM microservices, TensorRT-LLM, and Triton Inference Server. Includes Triton model configuration (config.pbtxt, dynamic batching, instance groups, version policies) and integrating LLMs with vLLM and Hugging Face Hub.
- Experimentation22%
Designing fine-tuning and alignment experiments with NeMo-Run recipes, including SteerLM's attribute-prediction, annotation, attribute-conditioned SFT, and bootstrap phases. Covers LM Evaluation Harness benchmarking, RAG evaluation, and FSDP2 distributed training.
- Data Analysis and Visualization14%
Preparing, profiling, and visualizing training and evaluation data for LLM workflows using RAPIDS (cuDF, cuML), Spark RAPIDS, and pandas. Covers dataset curation, tokenization stats, loss-masking inspection, and metrics dashboards for instruction-tuning and RAG quality.
Sources
Questions are grounded in 50 references from official and authoritative materials.