SnowPro Advanced Data Scientist

Question

A Data Scientist needs to build an LLM-based application in Snowflake that categorizes customer-support-call transcripts into predefined categories. A large labeled transcript data set is available. The categories are not expected to change over time, but inference must run on a very large number of transcripts. The top priority for the evaluation methodology is cost optimization, followed by accuracy.

What should the Data Scientist do to satisfy these requirements?

Explanation

For stable classification labels and a large labeled data set, fine-tuning avoids sending extensive labeled examples in every inference prompt, reducing recurring token consumption at high inference volume. Fine-tuning mistral-7b, the smaller supported base model, prioritizes lower cost while adapting the model to the classification task. Snowflake supports mistral-7b as a Cortex Fine-tuning base model and permits inference with the resulting fine-tuned model.

Learn more

SnowPro Advanced Data Scientist

AstroTutor

Notes

Readiness

QuestionQ33

Community Discussion