Chunking and Embedding Strategies GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 1)

Questions 1–4

1. 1foundation · easy

   During retrieval evaluation, a RAG system shows high precision (retrieved chunks are relevant) but low recall (many relevant chunks are missed). Which chunking strategy adjustment is most likely to improve recall?

   • AReduce the number of chunks retrieved per query
   • BReduce the chunk size to create more granular pieces
   • CSwitch to a different embedding model with the same context length
   • DIncrease the chunk size with some overlap between chunks
   Reveal Solution Discussion Select an answer first
2. 2foundation · easy

   An engineer must choose between two embedding models: Model X supports a 512-token context, and Model Y supports a 2,048-token context. The source documents are short (average 300 tokens), but the optimization strategy is to reduce the number of chunks per document to lower storage costs. Which model is more appropriate for this optimization?

   • AModel X, because shorter context models are always faster
   • BModel X, because it matches the document length exactly
   • CModel Y, because its larger context allows fewer, larger chunks per document
   • DModel Y, because larger context always produces better embeddings
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   A generative AI engineer is building a RAG pipeline and must decide on a chunking strategy. The team has run retrieval evaluations showing that smaller chunks improve retrieval precision but reduce the amount of context available to the model. Which consideration is most directly relevant when selecting the chunk size?

   • AThe number of GPUs used for training the embedding model
   • BThe tokenizer vocabulary size of the embedding model
   • CThe context window of the downstream language model
   • DThe embedding model's output dimensionality
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A team is selecting an embedding model for a RAG application. The source documents are long technical manuals (up to 8,000 tokens each), and queries are short (under 50 tokens). Which embedding model characteristic is most important to match to the source document length?

   • AThe embedding vector dimensionality
   • BThe maximum context length of the embedding model
   • CThe training data domain of the embedding model
   • DThe tokenizer's special token handling
   Reveal Solution Discussion Select an answer first