Top-k sampling restricts generation to:

t-SNE is limited because it:

ReLU differs from sigmoid because it:

Without positional encoding, Transformers:

PyTorch differs from NumPy because it:

In clustering workflows, the first step is:

A lookup by keyword or n-gram is an example of

A common failure of semantic search is:

Temperature in sampling controls:

What is the purpose of the softmax in attention?