Interview Practice Questions

Describe your implementation of an AI agent (a RAG system) in detail.

Сейчас. Ты говорил, что работал с VLM. Зачем вообще он нужен, если есть точку Generate с Transformers?

Why Word2vec, а не Berth какой-нибудь?

Можешь сказать, что из себя представляет эмбейдинг и Why он нам интересен в контексте языковой модели?

For example, a distilled BERT with a small output embedding and around 10–20 million parameters.

Then compare the cosine distance between our text and the product to determine how relevant the conversation is to the product.

How does the transformer account for token order?

What is the main difference between BERT and GPT?

How to aggregate transformer output into a single vector: CLS-token or pooling?

How do DPO, PPO and RLHF differ algorithmically?

What is PEFT and what approaches exist?

Tell about LLM in content moderation context

Tell about your experience with LLMs for chatbots

How can connecting another LLM provide additional information?

Did you use default LLMs with system prompts or fine-tuning for moderation?

How to approach the task of generated content detection?

Did you train LoRA adapter yourself or use a shared one for the department?

Мы сначала прогоняем через матрицу какую-то, да?

Инициализируется? Хороший вопрос. То есть какими значениями инициализируется?

Why do transformers use Layer Normalization instead of Batch Normalization?