Snowflake GES-C01 Exam : SnowPro® Specialty: Gen AI Certification Exam

A Gen AI developer is implementing a Cortex Search Service for a RAG application and needs to configure the text splitting for optimal performance using SNOWFLAKE.CORTEX.SPLIT_TEXT_RECURSIVE_CHARACTER Which of the following statements represent best practices or outcomes when applying text splitting with this function for Cortex Search in a RAG scenario? (Select all that apply)

• A. Even when using embedding models with larger context windows (e.g., 8000 tokens), a smaller chunk size is typically preferred for improved retrieval and downstream LLM response quality.
• B. Smaller chunk sizes generally lead to higher retrieval precision for a given query in a RAG system.
• C. Snowflake recommends splitting text into chunks of no more than 512 tokens for best search results in Cortex Search.
• D. Optimal text splitting using this function ensures that the number of input tokens precisely equals the number of output tokens for subsequent LLM calls, thereby minimizing compute costs.
• E. The function automatically enriches each text chunk with relevant metadata about its original document, such as author and creation date, for enhanced filtering capabilities in Cortex Search.

A Snowflake team observes consistently high token costs from 'SNOWFLAKE.ACCOUNT USAGE.CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY' for a summarization task using the 'mistral- large? model. The task involves summarizing legal documents, which often exceed the context window of common LLMs. To optimize these token-based costs, which strategy should the team prioritize?

• A. Implement a text splitting mechanism, potentially using SPLIT_TEXT_RECURSIVE_CHARACTER, to break down lengthy documents into smaller chunks before passing them to the summarization function, then aggregate the summaries.
• B. Set the temperature parameter to 0 in the COMPLETE function options to ensure more deterministic and thus more cost-efficient summarization outputs.
• C. Enable Cortex Guard for the COMPLETE function calls, as its filtering capabilities automatically reduce the number of tokens processed for unsafe content.
• D. Switch from using the COMPLETE function to TRY_COMPLETE to automatically avoid billing for queries that fail due to context window limits, thereby reducing costs.
• E. Increase the virtual warehouse size (e.g., from X-SMALL to MEDIUM) used for running the summarization queries to boost performance and reduce overall cost per query.

A data engineering team is designing a scalable data pipeline in Snowflake that involves processing large text inputs with Cortex AI LLM functions. They want to ensure cost efficiency and prevent queries from failing due to exceeding LLM context window limits. They plan to use SNOWFLAKE. CORTEX. COUNT_TOKENS for pre-validation. Which of the following statements are TRUE about the role and cost of COUNT_TOKENS in this scenario? (Select all that apply)

• A. Option C
• B. Option E
• C. Option A
• D. Option B
• E. Option D

A Snowflake Gen AI Specialist is defining a semantic model for Cortex Analyst to improve text-to-SQL accuracy. They are adding entries to the verified _ queries section of their YAML file. Consider the following semantic model snippet and a proposed verified_query entry. Which of the following statements correctly identifies an issue or a best practice not followed in the sql field of the proposed verified_query entry, based on Cortex Analyst VQR guidelines? Semantic Model Snippet:

Proposed verified _ query entry:

• A. Option C
• B. Option E
• C. Option A
• D. Option B
• E. Option D

• A.
  
• B.
  
• C.
  
• D. Cross-region inference is automatically managed by Snowflake for allowed models, implying that a new, larger virtual warehouse is required to handle the cross- region data transfer overhead.
• E.