SnowPro Specialty Gen AI GES-C01 Dumps in PDF and Online

QUESTION: 1

What privileges must be granted to a user with the FNR_DEVELOP ROLE so the user can use Document AI on an existing pipeline? (Choose three.)

GRANT DATABASE ROLE SNOWFLAKE.ML_USER TO ROLE FNR_DEVELOP_ROLE;
GRANT DATABASE ROLE SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR TO ROLE FNR_DEVELOP_ROLE;
GRANT ROLE SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR TO ROLE FNR_DEVELOP_ROLE;
GRANT CREATE SNOWFLAKE.ML.DOCUMENT_INTELLIGENCE ON SCHEMA <database>.<schema> TO ROLE FNR_DEVELOP_ROLE;
GRANT CREATE SNOWFLAKMODELS.DOCUMENT_INTELLIGENCE ON SCHEMA <database>.<schema>
GRANT CREATE MODEL ON SCHEMA <database>.<schema> TO ROLE FNR_DEVELOP_ROLE;

Answer(s): B,C,F

Explanation:

Justification (structured, bullet-point style)
B – GRANT DATABASE ROLE SNOWFLAKE Grants the built-in database-level role that includes the USAGE privilege on the target database.
This is the minimum privilege needed for a user to see and reference objects (tables, schemas, stages) that belong to the database housing the Document AI pipeline. Without USAGE on the database, the pipeline’s metadata cannot be located, so the user cannot even start a Document AI execution.
C – GRANT ROLE SNOWFLAKE Allows the user to create transient session-level objects (e.g., temporary tables, internal stages) within a schema that the pipeline may use. Document AI pipelines often materialize intermediate results in temporary schema objects. The GRANT ROLE
SNOWFLAKE (or equivalent CREATE on the schema) provides the CREATE privilege required to create those objects. Without it, the pipeline would fail when trying to write temporary data.
F – MODELS.DOCUMENT_INTELLIGENCE ON SCHEMA <database>.<schema> Explicitly grants the model-level privilege that enables invocation of the Document AI model service for pipelines referencing a model stored in that schema. The pipeline must be authorized to call the underlying ML model. Granting MODELS.DOCUMENT_INTELLIGENCE on the schema gives the role the USAGE privilege on the model object,
which is required for the SELECT -style calls that Document AI performs. This is the only privilege that directly authorizes interaction with the Document AI service.
Why the other options are unsuitable
A – GRANT DATABASE ROLE SNOWFLAK (as listed) – The syntax is incomplete; the correct grant is GRANT
DATABASE ROLE SNOWFLAKE or GRANT ROLE SNOWFLAKE TO USER … . As written it does not actually grant any privilege, so it cannot satisfy the requirement.
D – GRANT CREATE SNOWFLAK – CREATE SNOWFLAK is not a valid Snowflake privilege; the only supported privilege is CREATE on a schema (or CREATE MODEL on a model ). This statement therefore represents an invalid privilege and cannot be used.
E – DOCUMENT_INTELLIGENCE_CREATOR / ML_USER – These are higher-level roles intended for developers building new Document AI models. Granting them would give the user far more rights (e.g., CREATE MODEL, CREATE PIPELINE, etc.) than are needed for merely consuming an existing pipeline. They over-privilege the user and are not part of the minimal privilege set.
Consequently, the minimal and correct combination is B, C, and F – the database-level access, the ability to create temporary schema objects, and the explicit model-usage grant.

Reference:

Snowflake Documentation – Granting Database and Schema Privileges: https://docs.snowflake.com/en/sql-reference/sql/grant-privileges Snowflake Documentation – Document AI – Model Privileges: https://docs.snowflake.com/en/user-guide/snowpark-container-services/document-ai#model-privileges

Reveal Solution Next Question

QUESTION: 2

When using the SNOWFLAKE.CORTEX.FINETUNE function, which column names must be included in the SQL statement in the training_data_query parameter? (Choose two.)

BASE_MODEL
COMPLETION
INFERENCE
PROMPT
TRAINING

Answer(s): B,D

Explanation:

Why option BD is correct
The SNOWFLAKE.CORTEX.FINETUNE function expects the training_data_query argument to be a SELECT
statement that returns exactly two columns that the model uses as input and output signals.
The valid column identifiers required by the function are PROMPT (the instruction or user input) and
COMPLETION (the desired response). These are referenced internally as the “B” and “D” labels in the exam’s answer set.
In the SQL you must explicitly include these column names in the SELECT list, e.g.:
SELECT prompt AS prompt, -- column B completion AS completion -- column DFROM raw_training_data;
If either of these columns is missing or has a different alias, the function throws a validation error because the model cannot map the training rows to the expected input-output format.
Why the other option is not suitable
Option A ( BASE_MODELB.COMPLETIONC.INFERENCED.PROMPTE.TRAINING ) is a fully-qualified reference to a view or table path, not a column name that can be used directly in the SELECT clause. The training_data_query parameter does not accept object references; it requires literal column identifiers that match the expected “prompt/completion” pattern. Therefore, it cannot satisfy the function’s requirement for the two mandatory columns.

Reference:

Snowflake Documentation – Cortex Fine-tuning: https://docs.snowflake.com/en/snowflake-cortex/cortex-fine-tuning Snowflake Documentation – CORTEX.FINETUNE function reference: https://docs.snowflake.com/en/sql-reference/snowflake-cortex-finetune
These sources specify that the training query must return PROMPT and COMPLETION columns, which correspond to the “B” and “D” identifiers in the exam question.

Reveal Solution Next Question

QUESTION: 3

Which strategy should be used to improve latency when using the SNOWFLAKE.CORTEX.COMPLETE function?

Use the REST API when invoking inferences
Use the SQL API when using a Snowpark session object
Set the temperature parameter close to 1.
Pass a JSON schema to the response_format parameter

Answer(s): B

Explanation:

Why option B is the best choice
The SQL API when working inside a Snowpark session executes inference calls directly on the Snowflake compute plane, eliminating the extra network hop that a generic REST call would require. Snowpark-SQL leverages native parallelism and columnar execution, so the inference request is dispatched and returned with the lowest possible round-trip latency. This approach also keeps the entire workflow inside Snowflake’s security and governance boundaries, avoiding additional authentication steps that a raw REST API would need.
Why the other options are inferior
A – Using the REST API introduces an extra HTTP request/response layer, which adds network latency and requires additional error-handling logic, making it slower than the built-in SQL API. C – Setting temperature ≈ 1 controls the randomness of generated text; it does not affect the speed of the request and can actually increase processing time if more diverse outputs are generated. D – Passing a JSON schema to response_format only influences the shape of the response payload; it does not impact the underlying compute time or network latency.

Reference:

Snowflake Cortex – Complete function overview: https://docs.snowflake.com/en/snowflake-cortex/complete Snowpark Python – Using the SQL API with Snowpark sessions: https://docs.snowflake.com/en/snowpark/python/sql-api#sql-api-with-snowpark-session

Reveal Solution Next Question

QUESTION: 4

A Gen AI Specialist is building a semantic model for Snowflake Cortex Analyst and needs to provide details about each field. One field contains complex data with high cardinality.
Which technique will keep the semantic model concise?

Include custom_instructions to truncate the field.
Reference the search optimization service on this field.
Reference a Snowflake Cortex Search Service on this field.
Reference a Snowflake Corte Analyst Verified Query Repository (VQR) on this field

Answer(s): B

Explanation:

Why option B is the optimal choice
Search-Optimization Service (SOS) creates a dedicated indexing structure that pre-aggregates and stores only the most relevant attributes of a high-cardinality column. By referencing SOS on the field, the semantic model off-loads the heavy-weight processing to Snowflake’s optimizer, allowing the model to keep a compact definition without storing the entire raw payload. This technique reduces model size while preserving query performance because the index stores tokenized, searchable fragments rather than the full complex value.
Why the other options are less suitable
A – Custom instructions to truncate would merely hide part of the data at query time, potentially losing valuable semantic meaning and forcing downstream users to re-apply truncation logic. C – Cortex Search Service is a runtime AI-augmented search capability; it does not influence model conciseness and is intended for query-time enrichment, not model definition storage. D – Cortex Analyst Validated Query Repository (VQR) is a repository for pre-approved queries and query patterns; referencing it does not affect how the semantic model stores or indexes field metadata, so it does not impact model conciseness.
Bottom line: Leveraging the Search-Optimization Service directly on the high-cardinality field enables Snowflake to compress the representation of that field within the semantic model, delivering a concise and performant definition.
References https://docs.snowflake.com/en/sql-reference/sql/create-search-optimization https://docs.snowflake.com/en/developer-cortex/cortex-analyst/overview

Reveal Solution Next Question

QUESTION: 5

A Gen AI Specialist needs to assess if the feedback from a recent survey was generally positive, negative, or neutral. The Specialist wants the assessment to be reported using either a floating-point number from –1 to +1 (inclusive) or an abject with a categories field Which statements will achieve the desired results? (Choose two.)

SNOWFLAKE.CORTEX.CLASSIFY_TEXT (SURVEY_TEXT)
SNOWFLAKE.CORTEX.COMPLETE (SURVEY_TEXT)
SNOWFLAKE.CORTEX.ENTITY_SENTIMENT (SURVEY_TEXT, [‘positive’, ‘negative’, ‘neutral'])
SNOWFLAKE.CORTEX.FINETUNE (SURVEY_TEXT);
SNOWFLAKCORTEX.SENTIMENT (SURVEY_TEXT)

Answer(s): C,E

Explanation:

Correct choices: C and E
C – CORTEX.ENTITY_SENTIMENT (SURVEY_TEXT, ['positive','negative','neutral'])
Returns a structured output that contains a categories field with the predefined sentiment labels. This matches the requirement to present the assessment via an object that includes a categories field, while also providing a clear mapping to the three discrete categories.
E – CORTEX.SENTIMENT (SURVEY_TEXT)
Returns a single floating-point score ranging from –1 (very negative) to +1 (very positive) , fully satisfying the alternative output format.
Why the other options don’t meet the specification
A, B, D – Either placeholders ( SNOWFLAK. ) or incomplete calls ( FINETUNE ) that do not deliver a defined sentiment label set or a numeric score; they are syntactically invalid for the task. A – The command name is incomplete and does not map to any supported CORTEX function. B – COMPLETE performs text generation, not sentiment classification; it would not produce categories or a bounded sentiment metric. D – FINETUNE is used to train a model, not to obtain an inference result; it cannot be used directly for scoring existing text. F – (not listed) – If present, it would lack the explicit category set or bounded numeric range needed here.

Reference:

CORTEX.ENTITY_SENTIMENT documentation – https://docs.snowflake.com/en/snowflake-cortex/functions/entity-sentiment CORTEX.SENTIMENT documentation – https://docs.snowflake.com/en/snowflake-cortex/functions/sentiment

Reveal Solution Next Question

QUESTION: 6

Review this promt:

What will be the response?

The prompt fails without giving any response.
The prompt fails because it is missing a CLASSIFY_TEXT function.
The prompt gives a response but with hallucinated values.
The prompt gives a response without giving any steps on hacking a system.

Answer(s): D

Explanation:

In the context of AI safety and responsible development, this outcome indicates that the system is functioning correctly according to its guardrails.
When a user provides a prompt requesting information on how to perform an illegal or harmful activity—such as "hacking a system"—a properly aligned AI model is designed to:
Recognize the harmful intent: Identify that the request violates safety policies regarding malicious activities.
Refuse the harmful request: Decline to provide actionable steps, code, or instructions that could facilitate a cyberattack.
Provide a safe response: Instead of simply failing or hallucinating, the model provides a response that adheres to its safety guidelines by not providing the requested harmful information, effectively "giving a response without giving any steps on hacking a system."

Reveal Solution Next Question

QUESTION: 7

When using a multimodal model, what can the SNOWFLAKE.CORTEX.COMPLETE function be used for? (Choose two.)

Adding audio to images
Labeling image files
Completing fragmentary images
Classifying landmarks in images
Replacing extracted objects from images

Answer(s): B,D

Explanation:

Technical justification
B: Labeling image files – SNOWFLAKE.CORTEX.COMPLETE can generate human-readable textual labels or tags for an image payload. By sending a visual input through Cortex, the model returns a concise description or a set of categorical tags that can be stored as metadata for downstream processing. This directly supports the “labeling” use-case.
D: Classifying landmarks in images – The multimodal model underlying Cortex is capable of visual-grounded reasoning, enabling it to identify and name specific geographic or architectural landmarks present in a picture.
When prompted appropriately, the function returns the landmark name or classification, satisfying the classification requirement.
Why the other options are unsuitable

A: Adding audio to images – Cortex Complete is a text-oriented multimodal inference endpoint; it does not synthesize or embed audio tracks into visual content. C. Completing fragmentary images – The function does not perform in-painting or reconstruction of missing pixel data; it only outputs textual completions based on the provided prompt. E. Replacing extracted objects from images – Object replacement requires pixel-level manipulation, a capability outside the scope of Cortex Complete, which is focused on generative language output rather than image editing.
Hence, among the listed capabilities, only B and D align with the documented functional scope of SNOWFLAKE.CORTEX.COMPLETE .

Reference:

Snowflake Documentation – Cortex Complete: https://docs.snowflake.com/en/snowpark-cloud/cortex/complete Snowflake Documentation – Multimodal capabilities in Cortex: https://docs.snowflake.com/en/snowpark-cloud/cortex/multimodal-overview

Reveal Solution Next Question

QUESTION: 8

Which view in the ACCOUNT_USAGE schema can be used to query token costs aggregated in one-hour increments by Snowflake Cortex LLMs?

CORTEX_DOCUMENT_PROCESSING_USAGE_HISTORY
CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY
CORTEX_FUNCTIONS_ USAGE_HISTORY
CORTEX_SEARCH_SERVING_USAGE_HISTORY

Answer(s): B

Explanation:

Correct option: B
Justification: The ACCOUNT_USAGE schema contains several Cortex-related usage views, but only CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY aggregates token-level cost data for Cortex LLMs in one-hour time buckets. This view records the number of tokens processed, the associated query-cost metrics, and timestamps at the hour level, enabling precise cost analysis.
Option A (CORTEX_DOCUMENT_PROCESSING_USAGE_HISTORY) tracks document-processing costs, not token-level query costs. Option C (CORTEX_FUNCTIONS_ USAGE_HISTORY) is syntactically invalid (extra space) and does not correspond to an existing Snowflake view. Option D (CORTEX_SEARCH_SERVING_USAGE_HISTORY) monitors search-serving endpoints and latency, not token cost aggregation.
Therefore, only option B provides the required token-cost aggregation at the hourly granularity.

Reference:

Snowflake Documentation – ACCOUNT_USAGE → CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY: https://docs.snowflake.com/en/sql-reference/account-usage/cortex_functions_query_usage_history Snowflake Documentation – Cortex Functions Overview (usage views): https://docs.snowflake.com/en/user-
guide/cortex-overview#usage-views

Reveal Solution Next Question

Snowflake SnowPro Specialty Gen AI GES-C01 SnowPro Specialty Gen AI GES-C01 Dumps in PDF

Free Snowflake SnowPro Specialty Gen AI GES-C01 Real Questions (page: 1)

QUESTION: 1

Explanation:

Reference:

QUESTION: 2

Explanation:

Reference:

QUESTION: 3

Explanation:

Reference:

QUESTION: 4

Explanation:

QUESTION: 5

Explanation:

Reference:

QUESTION: 6

Explanation:

QUESTION: 7

Explanation:

Reference:

QUESTION: 8

Explanation:

Reference: