A data processing team is using Snowflake Document AI to extract data from incoming supplier invoices. They observe that many documents are failing to process, and successful extractions are taking longer than expected, leading to increased costs. Upon investigation, they find error messages such as

. Additionally, their 'X-LARGE virtual warehouse is constantly active, contributing to higher-than-anticipated bills. Which two of the following actions are essential steps to troubleshoot and address the root causes of these processing errors and optimize their Document AI pipeline?

• A. Implement a pre-processing step to split documents exceeding 125 pages or 50 MB into smaller, compliant files before loading to the stage.
• B. Increase the 'max_tokens' parameter within the '!PREDICT' function options to accommodate longer document responses from the model.
• C. Scale down the virtual warehouse to 'X-SMALL' or 'SMALL' size, as larger warehouses do not increase Document AI query processing speed and incur unnecessary costs.
• D. Redefine extraction questions to be more generic and encompassing, reducing the number of distinct questions needed per document.
• E. Configure the internal stage used for storing invoices with 'ENCRYPTION = (TYPE = 'SNOWFLAKE SSE')'.

A data scientist needs to generate vector embeddings for product descriptions stored in a column 'PRODUCT_DESCRIPTION' in the 'PRODUCT_CATALOG' table. They want to use the 'e5-base-v2 model for this task. Which of the following SQL statements correctly applies the 'SNOWFLAKE.CORTEX.EMBED TEXT 768' function and accurately describes the expected data type of the resulting embedding?

• A. The query
  
• B. The query
  
• C. The query
  
• D. The query
  
• E. The query
  

A developer is building a client application that interacts with a Snowflake Cortex Agent using its REST API. They are implementing multi- turn conversation support. Which of the following is the most critical aspect for maintaining conversational context over multiple API calls?

• A. The agent internally summarizes long conversations, and the client receives only a condensed summary_token to pass for subsequent turns.
• B. The REST API automatically identifies and reuses context from previous requests based on the Authorization token, eliminating the need to pass full history.
• C. Only the last user prompt and the immediately preceding agent response should be sent to conserve token usage.
• D. A specific session_id parameter, generated at the start of the conversation, must be sent with each request to link turns.
• E. The client application must include all prior user prompts and assistant/analyst responses in a messages array in each new API request.

Considering Snowflake's Gen AI principles for cost governance within Snowflake Cortex, an ML engineer is assessing the expenditure for an LLM fine-tuning job. Which option correctly identifies how compute costs for Cortex Fine-tuning are primarily incurred and how fine-tuned models are treated regarding usage by other customers?

• A. Only inference using fine-tuned models incurs costs, not the training itself. Fine-tuned models can be openly shared on the Snowflake Marketplace.
• B. Costs are based on the number of fine-tuning jobs created, not tokens. Fine-tuned models are shared across all Snowflake customers to improve the general service.
• C. Fine-tuning costs are a flat monthly fee, irrespective of token usage or model size. Fine-tuned models become part of Snowflake's proprietary models after training.
• D. Costs are incurred per hour of compute pool usage, similar to virtual warehouses. Fine-tuned models are anonymized and used to train future foundation models for all customers.
• E. Compute costs for fine-tuning are based on the number of tokens used in training, calculated as 'number of input tokens number of epochs trained'. Fine-tuned models built using a customer's data are available exclusively for that customer's use.