Week10_복습과제_최지희

[SweetFriedPotato]

No description provided.

[Copilot]

Pull request overview

This PR appears to add a Week 10 review assignment submission demonstrating KV-cache behavior (and its relationship to PagedAttention) via a small PyTorch attention prototype plus benchmark/visualization artifacts.

Changes:

• Add a Jupyter notebook implementing a simple “no cache vs KV-cache” attention computation and timing comparison.
• Add a PDF export of related materials/results.

Key review findings (changes requested):

• The notebook currently includes large embedded cell outputs (notably an image/png base64 blob around line 268) and captured stderr with local filesystem paths (around lines 256–263). This bloats the repo and can leak local environment details; clear outputs before committing (or use an output-stripping tool/workflow).
• Attention score scaling divides by sqrt(hidden_size) while num_heads/head_dim are defined but not used (e.g., around lines 110 and 161). Either implement per-head logic and scale by sqrt(head_dim), or remove the unused multi-head parameters to avoid implying multi-head attention.
• The KV-cache “append” uses torch.cat every step (around lines 157–158), which adds significant quadratic overhead due to repeated tensor reallocations/copies; for a fair benchmark, preallocate or accumulate in a list and concatenate once (or use indexed writes).

Reviewed changes

Copilot reviewed 1 out of 2 changed files in this pull request and generated no comments.

File	Description
Week10_복습과제_최지희.ipynb	KV-cache demo + timing/memory visualization for Week 10 review assignment
Week10_예습과제_최지희.pdf	PDF artifact accompanying the Week 10 materials

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.