Course file

sample_notes.txt

week10_rag/sample_notes.txt

Neural networks learn by adjusting weights to reduce loss. A gradient tells each parameter which direction to move so the output becomes less wrong over time.

Attention lets a model compare the current token to earlier tokens. It computes scores, turns them into weights, and uses those weights to mix information from the past.

Retrieval-augmented generation works by searching notes or documents first. The retrieved passages provide grounded context before an answer is written.

PyTorch automates gradient tracking with autograd. You still choose the model, loss function, optimizer, data, and evaluation process.