SnowPro Specialty: Gen AI · 10% of the exam

Snowpark Container Services for Gen AI: free practice questions

5 sample questions from our 25-question bank for this domain — answers and explanations included. These are the same scenario-based style as the real Snowflake exam.

1. A developer asks whether Snowflake will build their container image for them once a service specification references it. What is the correct answer?

  • A. No — the image must be built locally or in CI first, then pushed to the image repository; SPCS only runs pre-built images✓ Correct
  • B. Yes — SPCS automatically builds an image from a Dockerfile placed on a stage
  • C. Yes — the compute pool builds the image the first time the service starts
  • D. No — only Snowflake support can build and upload images on a customer's behalf
Explanation

The image repository is a registry to push pre-built images to, not a build service — you build the image yourself before pushing it. SPCS and compute pools run containers; they do not compile or build images.

2. A container in an SPCS service needs to read training files directly from a Snowflake stage without a separate data-transfer step. How is this typically accomplished?

  • A. By mounting the stage as a volume in the service specification✓ Correct
  • B. By copying the stage contents into a virtual warehouse first
  • C. By embedding the file bytes as environment variables in the containers section
  • D. By granting the compute pool a masking policy on the stage
Explanation

Volumes in the spec file can mount a Snowflake stage directly as a filesystem path, so the container reads and writes stage data with no extra transfer step. Warehouses don't hold container-readable files, and environment variables and masking policies are the wrong mechanism entirely.

3. Select two factors that indicate a workload should be hosted on Snowpark Container Services rather than a managed Cortex function.

  • A. The team needs to serve their own fine-tuned model weights that aren't a supported Cortex Fine-Tuning base model✓ Correct
  • B. The team needs a specific open-source model that isn't in the managed Cortex-hosted catalog✓ Correct
  • C. The team wants the simplest possible serverless SQL call with zero infrastructure to manage
  • D. The task is standard text classification already fully covered by an existing Cortex AI function
  • E. The team wants Snowflake to fully manage scaling with no compute pool to size
Explanation

Custom fine-tuned weights outside Cortex Fine-Tuning's supported base models, and a specific open-source model missing from the Cortex catalog, are the two classic SPCS triggers. Wanting zero infrastructure, a task already covered by an existing Cortex function, or fully automatic scaling all point toward staying on managed Cortex functions instead.

4. Which statement correctly contrasts billing between managed Cortex functions and Snowpark Container Services?

  • A. Managed Cortex functions consume serverless credits per call; SPCS is billed by compute pool uptime, including GPU pools✓ Correct
  • B. Both are billed identically per token processed, regardless of infrastructure
  • C. SPCS is always cheaper than managed Cortex functions regardless of workload
  • D. Managed Cortex functions require a running compute pool, exactly like SPCS
Explanation

Managed Cortex functions bill as serverless credit consumption per call, while SPCS bills for compute pool uptime whether or not it's actively processing requests, and GPU pools cost more per hour than CPU pools. Cortex functions run with no compute pool involved at all, so the two billing models aren't comparable on a flat cost basis.

5. A team needs a specific open-source vision-language model that is not part of Snowflake's managed Cortex model catalog, and all data must remain inside Snowflake. Which path fits?

  • A. Containerize the model and deploy it as a service on Snowpark Container Services✓ Correct
  • B. Call a managed Cortex AI function, since Cortex hosts every open-source model
  • C. Export the data to an external cloud GPU instance and call the model's API from there
  • D. Use AI_EXTRACT, since it can run arbitrary custom models
Explanation

SPCS is exactly for a model outside the managed Cortex catalog while keeping data in Snowflake's boundary. Managed Cortex doesn't host every open-source model, exporting data to an external instance breaks the in-boundary requirement, and AI_EXTRACT is a document field-extraction function, not a general model runtime.

20 more questions in this domain

Practice the full bank with instant grading, flashcards, and a timed mock exam.

Start practicing free
Snowpark Container Services for Gen AI — Free SnowPro Specialty: Gen AI Practice Questions | DataCertPrep — Certification Prep