Explanation:
Correct Answer: C
Why Option C is correct:
Why other options are incorrect:
Option A: Fixed-size chunking with sequential linking until reaching 200,000 tokens would still exceed the FM's context window limits. The approach of "linking multiple chunks sequentially" contradicts the problem statement about context window limitations.
Option B: While hierarchical chunking can be useful, using parent chunks of 8,000 tokens might still exceed some FM context windows. More importantly, this approach doesn't dynamically select only the most relevant content for each query.
Option D: Splitting documents into equal segments based on token count (80% of context window) doesn't respect semantic boundaries. Technical documents often have complex structures where splitting at arbitrary token counts could separate related concepts, leading to poor model performance. Additionally, processing segments independently before aggregation may lose cross-segment context.
Key Concepts:
Ultimate access to all questions.
No comments yet.
An enterprise application uses an Amazon Bedrock foundation model (FM) to process and analyze 50 to 200 pages of technical documents. Users are experiencing inconsistent responses and receiving truncated outputs when processing documents that exceed the FM's context window limits.
Which solution will resolve this problem?
A
Configure fixed-size chunking at 4,000 tokens for each chunk with 20% overlap. Use application-level logic to link multiple chunks sequentially until the FM's maximum context window of 200,000 tokens is reached before making inference calls.
B
Use hierarchical chunking with parent chunks of 8,000 tokens and child chunks of 2,000 tokens. Use Amazon Bedrock Knowledge Bases built-in retrieval to automatically select relevant parent chunks based on query context. Configure overlap tokens to maintain semantic continuity.
C
Use semantic chunking with a breakpoint percentile threshold of 95% and a buffer size of 3 sentences. Use the RetrieveAndGenerate API to dynamically select the most relevant chunks based on embedding similarity scores.
D
Create a pre-processing AWS Lambda function that analyzes document token count by using the FM's tokenizer. Configure the Lambda function to split documents into equal segments that fit within 80% of the context window. Configure the Lambda function to process each segment independently before aggregating the results.