NVIDIA Certified Associate Generative AI LLMs (NCA-GENL) Practice Questions and Study Guide

In a transformer, what does the self attention mechanism let the model do?

• AProcess tokens strictly one at a time, passing a hidden state from each token to the next one
• BWeigh how relevant every other token in the sequence is when building the representation of each token
• CSplit the input text into subword units before the text is turned into embedding vectors
• DReduce the number of model parameters by sharing one weight matrix across every layer

A regression model has an R squared of 0.82 on the test set. What does this tell you?

• AThe model predicts the exact value correctly for 82% of test rows
• BThe average error of the model is 0.82 units of the target value
• CThe model will be 82% accurate on any future dataset it receives
• DThe model explains about 82% of the variance in the target values

What does LoRA (low rank adaptation) do when fine tuning an LLM?

• AIt freezes the base weights and trains small low rank update matrices
• BIt retrains every weight in the model using a much larger dataset
• CIt removes whole layers from the model to make inference run faster
• DIt replaces the tokenizer with a smaller vocabulary for the new domain

Your team must serve a PyTorch model, an ONNX model and a TensorRT engine behind one inference service with HTTP and gRPC endpoints. Which NVIDIA tool fits best?

• ARAPIDS cuML running each model inside a separate Python notebook
• BNVIDIA NeMo Curator configured to load all three model formats
• CNVIDIA Triton Inference Server with a backend for each framework
• DNCCL configured to send requests between the three model files

A bank chatbot must stay on banking topics and refuse to give medical advice, even when users try to steer it. Which NVIDIA open source toolkit is built to add these programmable controls?

• ARAPIDS cuDF
• BNeMo Guardrails
• CNVIDIA TensorRT-LLM
• DNVIDIA NCCL