What was used to build the index in RAG?
Machine Learning / AI
What to do at the end of training?
What is the difference between data parallelism and model parallelism?
How do you output a list of results?
How does ROC-AUC differ from other metrics?
What is sparse attention?
What is the purpose of the anchor in Triplet Loss?
What is the difference between Margin Loss and Triplet Loss?
What is an epoch?
What is n in the context of attention complexity?
How is a reranker trained?
If there are no reference answers, only documents and context — how to evaluate the system and where to get the reference answer?
What types of optimization exist for LLM?
What else besides cross-entropy can be used as a loss function?
What has changed in LLM compared to the original transformer?
What type of attention allows reducing complexity below O(n²)?
What is gradient checkpointing?
What is quantization?
How to evaluate an agent implementing RAG?
What happens if you input a sequence of 10k tokens to a decoder, but the model was trained on 9k?