NVIDIA·Machine Learning Engineer·Technical Phone Screen
- Walk through the architecture of a Transformer-based large language model. How does the self-attention mechanism allow the model to capture dependencies between distant tokens?
- You have a pretrained LLM and want to fine-tune it on a domain-specific dataset. How do you approach this without causing catastrophic forgetting?
- For an LLM deployed in production to answer user queries, how would you detect, monitor, and reduce hallucinations?
“I started with the encoder-decoder framing and then pivoted to decoder-only since that's what most LLMs actually use now.”