Certified Generative AI Engineer AssociateDemo

Question

A Generative Al Engineer is assigned to build a RAG application that helps a small internal group of company experts answer specific questions, augmented with an internal knowledge base. They require the highest possible answer quality; latency and throughput are not major concerns because the user group is small and willing to wait for the best answer. The subjects are sensitive and the data is highly confidential, so regulatory requirements prohibit transmitting any information to third parties.

Which model satisfies all of the Generative Al Engineer’s needs in this scenario?

A Dolly 1.5B
B OpenAI GPT-4
C BGE-large
D Llama2-70B

Explanation

Llama2-70B is a large generative language model that can be deployed within the company’s own environment, so the RAG prompts and retrieved confidential content need not be sent to a third-party model provider. Its 70-billion-parameter size makes it the strongest answer-generation choice listed when latency and infrastructure cost are acceptable. BGE-large is intended for embeddings and retrieval, not generation; a 1.5B Dolly model is substantially smaller; and GPT-4 requires sending requests to OpenAI’s service. Meta’s Llama 2 model card identifies 70B as a Llama 2 variant and describes the family as generative text models.

Learn more

Certified Generative AI Engineer AssociateDemo

AstroTutor

Notes

Readiness

QuestionQ72

Community Discussion