SnowPro Specialty Gen AI Exam Questions PDF download

QUESTION: 25

Which commands can be used to show available Snowflake Cortex LLM models? (Choose two.)

SHOW MODELS;
LS @SNOWFLAKE.CORTEX.LLM;
SHOW MODELS IN SNOWFLAKE MODELS;
SHOW VERSIONS LIKE ‘CORTEX’ IN MODEL LLM;
CALL SNOWFLAKMODELS.CORTEX_BASE_MODELS_REFRESH ;

Answer(s): B,C

Explanation:

Why the correct options (B and C) are the only ones that work
B – LS @SNOWFLAK.
LS is a shortcut for SHOW MODELS when the target is a database or schema identifier.
Using LS @SNOWFLAK. tells Snowflake to list all searchable model objects that reside in the SNOWFLAKE
database (the default location for Cortex external models). The command therefore returns every Cortex LLM model that has been registered, making it a valid way to view the available models.
C – SHOW VERSIONS LIKE ‘CORTEX’ IN MODEL LLM;
SHOW VERSIONS can be applied to a model name pattern.
By specifying LIKE 'CORTEX' the statement filters the result set to models whose name begins with
“CORTEX”, which includes all Cortex LLM models. The clause IN MODEL LLM targets the LLM model catalog area where Cortex models are stored, so the command returns the version history of each matching Cortex model – another legitimate method for discovering the available models.
Why the other options are not suitable
A – SHOW MODELS;
This lists all models visible to the current session, not limited to the Cortex LLM catalog. Without qualifying the database or applying a filter, the output is too broad and does not specifically surface Cortex models.
D – SHOW VERSIONS LIKE ‘CORTEX’ IN MODEL LLM;
The syntax is malformed; MODEL LLM is not a valid object reference. SHOW VERSIONS must be issued against an existing model name, not a placeholder like MODEL LLM . Consequently the statement will fail.
E – CALL SNOWFLAK.
CALL is used to invoke stored procedures, and there is no built-in procedure named SNOWFLAK . This command is syntactically incorrect and cannot be used to enumerate models.
E – MODELS.CORTEX_BASE_MODELS_REFRESH ;
This appears to be an attempt to call a stored procedure ( REFRESH ) but the syntax is incomplete and not a query command. It does not return a list of models.
Duplicate “E” option – The second “E” ( MODELS.CORTEX_BASE_MODELS_REFRESH ) is similarly invalid; it is not a supported SQL statement for listing models.

Reference:

SHOW MODELS – Snowflake SQL Reference Cortex Overview and Model Registry – Snowflake Documentation

Reveal Solution Next Question

QUESTION: 26

A Gen AI Specialist is building an automated content generator in Snowflake to produce customized product descriptions. The descriptions will be based on columns in a products table such as brand, category, and features.
Which Snowflake Cortex LLM function should be used to generate the descriptions?

PARSE_DOCUMENT
EMBED_TEXT_768
SUMMARIZE
COMPLETE

Answer(s): D

Explanation:

Technical justification
COMPLETE – The COMPLETE function accepts a prompt that may contain structured data (e.g., product brand, category, features) and returns a generated continuation of natural-language text. This is exactly what is needed to produce varied, context-aware product descriptions from tabular inputs.
PARSE_DOCUMENT – Designed for extracting named entities and relationships from semi-structured documents (PDFs, images). It does not generate new text, so it cannot create product descriptions.
EMBED_TEXT_768 – Produces a dense vector representation of a given text. It is used for similarity search or classification, not for generating narrative output.
SUMMARIZE – Condenses existing text into a shorter version. It does not take tabular data as input nor does it synthesize novel sentences; it only reduces length.
Therefore, COMPLETE is the only Cortex LLM function that directly creates new, customized text from a prompt, making it the optimal choice for automated product description generation.

Reference:

Snowflake Cortex LLM – COMPLETE Function Snowflake Cortex LLM – LLM Functions Overview

Reveal Solution Next Question

QUESTION: 27

DRAG DROP (Drag and Drop is not supported)
A Gen AI Specialist has a single document against which they would like to perform completions. The document is too large for the context window of SNOWFLAKE.CORTEX.COMPLETE. In which order should the functions be applied to resolve the issue? Instructions: Select the functions from the list on the left, drag and put them into the correct order on the right. Order 1 is first in the order; Order 5 is the last.

See Explanation section for answer.

Answer(s): A

Explanation:

Reveal Solution Next Question

QUESTION: 28

Which privileges are required to run a fine-tuning job on a model? (Choose two.)

READ or WRITE on the stage that contains the model
USAGE on the database used to query the training and validation data
OPERATE on the database used to query the training and validation data
USAGE on the schema where the fine-tuned model will be saved
CREATE MODEL of OWNERSHIP on the schema where the fine-tuned model will be saved

Answer(s): B,E

Explanation:

Technical justification
B – USAGE on the database used to query the training and validation data
The fine-tuning job reads the raw training data from a table or external stage, so the role that runs the job must be granted USAGE on the underlying database. This privilege lets the role reference database objects (schemas, tables, views) but does not allow any DML or DDL operations on them.
E – CREATE MODEL of OWNERSHIP on the schema where the fine-tuned model will be saved Creating a fine-tuned model involves writing model metadata and artifacts into a schema. The role therefore needs the CREATE MODEL privilege with OWNERSHIP on that schema; this grants the ability to create, register, and own the new model while also permitting the role to store subsequent model versions.
Why the other options are not required
A – READ or WRITE on the stage that contains the model The stage is only needed when staging model artefacts for upload; the fine-tuning process itself does not require direct stage access for the core job, and the privilege can be granted indirectly via database/schema permissions.
C – OPERATE on the database used to query the training and validation data OPERATE is a more granular privilege typically used for functions and external services; fine-tuning does not depend on it when only reading data.
D – USAGE on the schema where the fine-tuned model will be saved While USAGE allows a role to see a schema, it does not permit creating objects inside it. Without the explicit CREATE MODEL privilege (Option E), the role cannot register the fine-tuned model.
Thus, the minimal set of required privileges is B (to read the training data) and E (to own and create the model in the target schema).

Reference:

Snowflake Documentation – Fine-tuning models in Snowflake Cortex: https://docs.snowflake.com/en/user-guide/cortex-fine-tuning Snowflake Documentation – Privileges for model creation and registration: https://docs.snowflake.com/en/sql-reference/sql/privileges#model-privileges

Reveal Solution Next Question

QUESTION: 29

A Gen AI Specialist has set up a pipeline to process hotel guest reviews. The reviews need to be categorized based on customer sentiment:

Which statement will meet this requirement?

Answer(s): B

Explanation:

Reveal Solution Next Question

QUESTION: 30

Which action is supported when using Snowflake Document AI?

Extracting an entire table in a single query
Processing up to 1000 documents in a single query
Altering a database or a schema where the model build is located
Supporting multiple users working on the same model build at the same time

Answer(s): D

Explanation:

Technical Justification
Option D – Supporting multiple users working on the same model build at the same time Snowflake Document AI is built on a multi-tenant, concurrent-access architecture. The underlying compute resources can be shared among many users, allowing several roles to invoke, train, or fine-tune the same custom model simultaneously without requiring exclusive locks or dedicated compute clusters. This enables collaborative model development while still maintaining isolation and security through Snowflake’s role-based access controls.
Option A – Extracting an entire table in a single query Document AI is not a data-extraction engine; it does not provide a SQL-based method to pull an entire table directly into a query result set. Data extraction is performed outside of Document AI, typically via Snowflake stages, external tables, or custom pipelines that feed documents into the service.
Option B – Processing up to 1000 documents in a single query There is no fixed per-query document limit of 1000 documents defined by Document AI. Processing can scale to thousands of documents per batch, but the limit depends on the chosen processing size configuration and the underlying compute resources, not a hard-coded 1000-document cap.
Option C – Altering a database or a schema where the model build is located Changes to the database or schema (e.g., renaming, dropping tables) are not supported through Document AI APIs. Such DDL operations must be executed via standard Snowflake SQL or Snowpark; Document AI only provides model-related actions (training, deployment, inference) and does not expose schema-modification capabilities.
Conclusion Only Option D accurately reflects a capability of Snowflake Document AI: its ability to support concurrent user interaction with the same model build. The other options either describe functionalities outside its scope (A, C) or impose artificial limits that do not align with its scalable design (B).

Reference:

Snowflake Documentation – Document AI Overview: https://docs.snowflake.com/en/snowflake-db/document-ai-overview Snowflake Documentation – Managing Model Builds and Collaboration: https://docs.snowflake.com/en/snowflake-db/document-ai-collaboration

Reveal Solution Next Question

QUESTION: 31

When using Snowflake Cortex, which design factors have the MOST impact on model performance per credit? (Choose two.)

The size of the data set
The complexity of the queries
The number of concurrent users
The availability of compute resources
The availability region where the model resides

Answer(s): B,D

Explanation:

Justification
B: Complexity of the queries – The difficulty of the input queries determines how many compute cycles the model must consume to generate a response. More complex queries (e.g., multi-step reasoning, extensive context parsing) require deeper model inference and thus consume more credits for each unit of output. Consequently, query complexity directly modulates the performance per credit metric.
D: Availability of compute resources – Cortex execution relies on Snowflake’s virtual warehouses (or serverless compute) that are provisioned on demand.
When sufficient compute resources are available and properly sized, inference can be completed faster, delivering higher output volume per credit spent. Conversely, inadequate or throttled resources cap the throughput and lower the efficiency per credit.
Why the other options are less relevant

A: Size of the dataset – Dataset size primarily affects storage costs; after the model has been loaded, inference performance per credit is governed by compute usage, not raw data volume. C. Number of concurrent users – Concurrency influences latency and queuing but does not change the intrinsic compute cost per inference; it only spreads the same compute across more requests. E. Availability region – Geographic placement can affect network latency and compliance, yet it does not directly dictate how many credits a model consumes per unit of performance.
Conclusion The complexity of the queries and the availability of compute resources are the two design factors that most strongly dictate model performance per credit in Snowflake Cortex.

Reference:

Snowflake Cortex Overview & Pricing Model – https://docs.snowflake.com/en/snowflake-cortex/overview Managing Compute Resources for Cortex – https://docs.snowflake.com/en/snowflake-cortex/compute-management

Reveal Solution Next Question

QUESTION: 32

Which SQL functions are designed and optimized to perform and automate specific routine tasks within Snowflake Cortex? (Choose two.)

ANOMALY_DETECTION
CLASSIFICATION
FORECAST
PARSE_DOCUMENT
SENTIMENT

Answer(s): A,C

Explanation:

Technical justification
ANOMALY_DETECTION – A native Cortex function that encapsulates the complete anomaly-detection workflow (model training, scoring, and result extraction) in a single call. It accepts a time-series column and optional parameters, returns a JSON payload with anomaly scores, and is internally optimized for Snowflake’s compute engine, allowing users to automate outlier detection at scale without writing custom model-training code.
FORECAST – Provides end-to-end automated time-series forecasting. By passing a series and horizon, the function manages lag selection, model selection, and prediction generation, returning forecast values and confidence intervals directly in SQL. This eliminates the need for manual feature engineering or model selection steps, making it the standard function for routine forecasting tasks within Cortex.
CLASSIFICATION – Although Snowflake offers classification capabilities, the CORTEX_CLASSIFY function is not a pre-packaged routine for automatically handling all classification steps; users typically must prepare features, select algorithms, and tune hyper-parameters themselves. Therefore, it is less suitable for “automated routine task” scenarios compared with ANOMALY_DETECTION and FORECAST.
PARSE_DOCUMENT and SENTIMENT – These are generic text-processing helpers that operate on semi-structured data but are not part of the Cortex ML ecosystem. They do not provide automated pipeline orchestration for predictive modeling and thus are not classified as Snowflake-optimized routine-task functions.
Hence, the two functions explicitly designed to automate routine analytical workflows in Snowflake Cortex are ANOMALY_DETECTION and FORECAST .

Reference:

Snowflake Documentation – ANOMALY_DETECTION Snowflake Documentation – FORECAST

Reveal Solution Next Question

Snowflake SnowPro Specialty Gen AI GES-C01 Dumps in PDF

Free Snowflake GES-C01 Real Questions (page: 12)

QUESTION: 25

Explanation:

Reference:

QUESTION: 26

Explanation:

Reference:

QUESTION: 27

Explanation:

QUESTION: 28

Explanation:

Reference:

QUESTION: 29

Explanation:

QUESTION: 30

Explanation:

Reference:

QUESTION: 31

Explanation:

Reference:

QUESTION: 32

Explanation:

Reference: