1Z0-1127-25 Practice Test 50 — 15 Questions

Question 1

An AI engineer runs a RAG chatbot on OCI Generative AI. Users report that answers drift off-topic when a single query retrieves many loosely related chunks. The engineer wants to reduce context noise while keeping the most semantically relevant passages. Which approach best addresses this?

Choose an answer to begin — your selection is scored in the full session.

Accepted Answer

Apply a similarity score threshold and rerank the retrieved chunks, then pass only the top few to the model.. Off-topic drift from loosely related chunks is a context-quality problem. Applying a similarity threshold discards weak matches and a reranker reorders survivors by relevance to the query, so only the strongest passages reach the model. That reduces noise and keeps the answer grounded, whereas enlarging top_k, changing embedding dimension, or lowering temperature does not improve which chunks are selected.

Answer

Increase the top_k of the vector search so more candidate chunks are passed to the model for grounding.

Answer

Lower the model's temperature to zero so the LLM produces more deterministic answers from the same context.

Answer

Switch the embedding model to one with a larger output dimension so each chunk carries more semantic detail.

Question 2

A solutions architect is designing an OCI Generative AI application that must respond to user prompts within 1 second. The architect is evaluating the Llama 3.1 405B model but is concerned about inference latency. Which factor most directly affects the model's inference speed during generation?

Accepted Answer

The number of parameters in the model. Inference latency is primarily determined by the model's parameter count, as each generated token requires a forward pass through all parameters. Larger models like 405B are slower than smaller ones. While other factors like hardware and batching matter, the parameter count is the fundamental driver of per-token computation time.

Answer

The tokenizer's vocabulary size

Answer

The size of the training dataset used to pretrain the model

Answer

The temperature setting used during inference

Question 3

A data scientist is preparing a dataset to fine-tune a foundation model on OCI Generative AI for a domain-specific Q&A task. They want to reduce the risk of the model hallucinating outside the provided domain knowledge. Which two data preparation practices are most appropriate? (Choose two.)

Accepted Answer

Add explicit examples where the correct response is 'I don't know' for out-of-scope questions.. Reducing hallucination in a fine-tuned model requires teaching it both the domain knowledge and the boundaries of that knowledge. Diverse, representative examples ensure the model learns the full scope of valid questions, while explicit 'I don't know' examples give it a learned abstention behavior for out-of-scope inputs. Together these practices improve reliability. Unlabeled text, easy-only filtering, and terminology removal do not provide the corrective signal needed.

Answer

Maximize dataset size by including as much unlabeled web text as possible.

Answer

Use only questions that the model already answers correctly to avoid confusing it.

Answer

Remove all examples that contain domain-specific terminology to simplify the text.

Question 4

A financial analyst is using an OCI Generative AI model to generate quarterly earnings summaries. The model occasionally produces summaries that include fabricated financial figures not present in the source data. Which phenomenon is primarily responsible for this behavior?

Accepted Answer

Hallucination. Hallucination is the correct term for when an LLM generates factually incorrect or fabricated information that is not supported by the input data. In this scenario, the model invents financial figures, which is a hallmark of hallucination. Overfitting, catastrophic forgetting, and vanishing gradients are training-related issues that do not directly cause inference-time fabrication.

Answer

Catastrophic forgetting

Answer

Vanishing gradients

Answer

Overfitting

Question 5

A team's LangChain RAG application uses a vector store retriever configured with search_type="mmr". Users complain that answers omit important details that appear in the top-scoring document. Which change best addresses this while preserving result diversity?

Accepted Answer

Increase the fetch_k parameter so MMR selects from a larger candidate pool before reranking for diversity.. Maximal Marginal Relevance retrieves an initial candidate set of size fetch_k, then greedily selects documents that balance query similarity with dissimilarity from already-selected results. When fetch_k is too small, relevant documents can be filtered out before the diversity pass. Increasing fetch_k gives MMR a richer pool so important top-scoring content survives selection while redundancy is still reduced.

Answer

Increase the chunk_overlap in the text splitter so retrieved chunks share more content with neighboring chunks.

Answer

Reduce the k parameter so that fewer documents are returned and each answer focuses on the single best match.

Answer

Switch search_type to "similarity_score_threshold" and set a low score threshold to include more documents.

Question 6

A solutions architect is deploying a customer-facing application on OCI that uses the OCI Generative AI Service. Security policy requires that all API traffic to the service never traverse the public internet. The application runs on compute instances in a private subnet of a VCN. Which capability should the architect use to satisfy this requirement?

Accepted Answer

Configure a service gateway in the VCN and create a private endpoint for the OCI Generative AI Service.. Private connectivity to OCI Generative AI is achieved by combining a service gateway in the VCN with a private endpoint for the service. This keeps inference traffic on the Oracle network, avoiding the public internet entirely, and is the supported pattern for workloads that must not egress publicly. Other connectivity options either route over the internet or connect to on-premises networks rather than to the service itself.

Answer

Enable a NAT gateway in the VCN so outbound calls from the private subnet are translated and remain invisible to external observers.

Answer

Use a dynamic routing gateway with a Site-to-Site VPN back to the on-premises data center, then proxy Generative AI calls from there.

Answer

Attach an internet gateway to the private subnet and restrict access with a network security group rule that allows only the Generative AI service CIDR.

Question 7

A solutions architect is designing a RAG pipeline on OCI Generative AI for a legal research assistant. The knowledge base contains 2,000-page contracts, and the assistant must answer clause-level questions with precise citations. Which chunking strategy should the architect implement to maximize retrieval precision and keep the generation grounded?

Accepted Answer

Split each contract into overlapping chunks aligned to semantic section boundaries, embed each chunk, and retrieve only the top-scoring chunks that fit the context window.. Semantic chunking with overlap aligned to document structure preserves clause meaning across boundaries, and top-k retrieval restricts the prompt to the most relevant passages. This combination keeps inputs within the context window while grounding generation in exact contract text, which is what enables precise clause-level citations and minimizes hallucinated legal content.

Answer

Embed the entire 2,000-page contract as one vector and pass the full document to the model on every query.

Answer

Fine-tune the base model on the full contract corpus and skip retrieval entirely for each query.

Answer

Chunk the contract into fixed 128-token segments with no overlap and retrieve a single best-matching segment.

Question 8

A developer is building an OCI Generative AI application that requires the model to answer questions based on a large, dynamic knowledge base. The knowledge base is updated frequently, and the developer wants to avoid retraining the model. Which approach is most suitable?

Accepted Answer

Use retrieval-augmented generation (RAG) with a vector database. Retrieval-augmented generation (RAG) combines an LLM with a retrieval system, allowing the model to access up-to-date information from a vector database without retraining. This is efficient for dynamic knowledge bases and ensures answers are grounded in current data. Fine-tuning or larger models do not provide real-time updates.

Answer

Increase the model's temperature to encourage creativity

Answer

Fine-tune the model on the entire knowledge base daily

Answer

Use a larger model with more parameters

Question 9

A developer is instrumenting a LangChain application that calls OCI Generative AI models so that prompts, completions, token usage, and latency are captured for troubleshooting and cost analysis. Which TWO LangChain facilities should they use? (Choose two.)

Accepted Answer

LangSmith tracing, by setting environment variables so runs are recorded and inspected in the LangSmith UI.. Callbacks provide programmatic event hooks that expose prompts, completions, errors, and timing for each LLM and tool invocation, while LangSmith tracing records runs with inputs, outputs, token usage, and durations in an inspectable UI. Together they deliver the observability needed for troubleshooting and cost analysis without altering application logic.

Answer

The ConversationBufferMemory class, by storing prior turns alongside token counts.

Answer

The VectorStore retriever, by logging the documents returned for each query.

Answer

Pydantic output parsers, by defining a schema that validates the model's response structure.

Question 10

An ML engineer is building a semantic search feature over 50 million internal knowledge articles using OCI Generative AI embeddings. Queries must return the most relevant passages, and the engineer must choose the representation the model produces for each passage. Which statement correctly describes what the embedding model outputs and how it is used?

Accepted Answer

A dense fixed-length vector whose geometric proximity to the query vector reflects semantic similarity. Embedding endpoints return one dense, fixed-length vector per input text, trained so that semantically similar texts land close together in vector space. Semantic search embeds the query, then ranks stored passage vectors by cosine or dot-product similarity through an approximate nearest-neighbor index, which is what makes retrieval over tens of millions of articles tractable.

Answer

A sequence of per-token hidden states that must be compared token-by-token between query and passage

Answer

A probability distribution over the vocabulary that is compared using Kullback-Leibler divergence

Answer

A sparse bag-of-words vector whose dimensions are the vocabulary terms, so exact keyword overlap drives ranking

Question 11

A team is evaluating a large language model for an OCI Generative AI deployment and wants to understand why the model can continue a passage fluently even though it has no database of verified facts. Which two statements correctly explain the pretraining objective and its consequence? (Choose two.)

Accepted Answer

The model is trained to predict the next token given preceding context, so it learns statistical patterns of language rather than storing verified facts. Pretraining with next-token prediction optimizes the likelihood of the next token, teaching language structure and distributing factual associations implicitly across the weights. There is no truth oracle or versioning, so facts can be stale or wrong while the surface text stays fluent. Retrieval, citations, and truth-oriented rewards are separate additions, not inherent properties of the base model.

Answer

The model is trained with a reinforcement learning reward that penalizes every factually false statement before deployment

Answer

The model memorizes each training document verbatim and can retrieve the exact source text on demand

Answer

The model queries an external knowledge graph at inference time and cites the retrieved node for every generated sentence

Question 12

A machine learning engineer is serving a fine-tuned OCI Generative AI model for an internal knowledge assistant. Users report that identical prompts return noticeably different answers between requests, and a compliance reviewer requires deterministic output for audit replay. Which inference configuration should the engineer apply to make responses reproducible?

Accepted Answer

Set temperature to 0 and disable top_p and top_k sampling so the model always selects the highest-probability token.. Reproducibility requires removing every stochastic element from decoding. Setting temperature to zero turns sampling into greedy selection of the most probable token, and disabling top_p and top_k prevents nucleus or top-k filtering from reintroducing randomness. With those settings, identical prompts and parameters yield identical completions, satisfying the audit replay requirement.

Answer

Enable beam search with a large beam width and leave temperature at its default.

Answer

Set temperature to a very high value and increase top_p to 1.0.

Answer

Increase max_tokens and set frequency_penalty to a negative value.

Question 13

A developer is building an OCI Generative AI application that must translate short product descriptions from English into Spanish. The descriptions are one or two sentences, and the developer wants the highest translation quality without writing training data. Which approach should the developer take?

Accepted Answer

Use prompt engineering with a pretrained instruction-following model, providing the source text and an explicit instruction to translate into Spanish.. A pretrained instruction-following model already handles multilingual generation, so a prompt that supplies the source description and explicitly requests Spanish translation delivers strong quality with zero training data and no infrastructure to manage. This matches the developer's goal of high quality with minimal effort for short texts.

Answer

Fine-tune a base model on the company's existing English product catalog before translating.

Answer

Deploy an embedding model and compute cosine similarity between English and Spanish product vectors.

Answer

Train a custom model from scratch on a parallel English-Spanish corpus using OCI Data Science.

Question 14

A developer is assembling a LangChain pipeline that must retrieve documents from an OracleVS vector store and then answer questions with an OCI Generative AI chat model. They want the fewest moving parts and no custom prompt engineering for the final answer step. Which LangChain component should they use to combine retrieval and generation in one object?

Accepted Answer

A RetrievalQA chain constructed with the vector store retriever and the OCI Generative AI chat model.. RetrievalQA bundles a retriever and a language model behind a default question-answering prompt, so a developer can point it at an OracleVS retriever and an OCI Generative AI chat model and get end-to-end retrieval-augmented answers with minimal configuration. The other components either require custom prompts, omit retrieval, or serve unrelated summarization or memory roles.

Answer

A MapReduceDocumentsChain that maps over each retrieved document and reduces the outputs.

Answer

A StuffDocumentsChain with a custom prompt and no retriever.

Answer

A ConversationBufferMemory object attached directly to the vector store.

Question 15

A developer is instrumenting a LangChain application that calls OCI Generative AI models. They need to trace each chain step, inspect the exact prompt sent to the model, and measure per-step latency in a hosted dashboard for debugging in staging. (Choose two.)

Accepted Answer

Enable LangSmith tracing by setting the appropriate LangChain tracing environment variables and API key.. LangChain provides two supported observability paths: hosted tracing through LangSmith, enabled with tracing environment variables and an API key, and programmatic instrumentation through callback handlers that receive lifecycle events like on_llm_start and on_llm_end. Both expose prompts, chain steps, and timing, while generation parameters and storage choices have no bearing on tracing.

Answer

Increase the model's temperature setting so the service returns more diagnostic metadata with each response.

Answer

Switch the vector store from OracleVS to an in-memory store so all operations are visible in application logs.

Answer

Raise the max_tokens parameter on the OCI Generative AI model so full prompts are echoed in the response.