How to use BERT in a SQuAD-style QA task?
Machine Learning / AI
What features are used in classic NLP models (TF-IDF, n-grams, character n-grams)?
How to reduce the size of KV-cache (GQA, MQA, sliding window)?
What tasks does Deepgram solve and what are its API features?
What are the practical problems of RLHF (reward hacking, mode collapse)?
How to choose the quantization format for inference on CPU vs GPU?
What is sampled softmax and what is its advantage in terms of compute?
Что такое EfficientDet?
What hyperparameters of HNSW are important (M, efConstruction, efSearch)?
How does sparse retrieval (BM25) differ from dense?
What streaming inference techniques for STT (chunked attention, look-ahead)?
What are class activation maps (CAM) and Grad-CAM?
What are DeepSORT and ByteTrack for multi-object tracking?
Compare classification, localization, and detection.
What are the features of real-time video detection (TensorRT, batch=1)?