∗ note
Al
cu
ı
n
Writing
Courses
Watch
Problems
coming soon
Labs
coming soon
Papers
coming soon
Minibook
coming soon
Sign in
EARLY ACCESS
☼
WED · NOV 13 · 09:42
Welcome back,
Sai.
You left off in the middle of a derivation. Let's
finish it.
STREAK
23
days
SOLVED
147
problems
RESUME · CH. 14 — ATTENTION
p. 142 / 178
Why we divide
by
√d
k
.
You stopped mid-paragraph. The variance argument resolves on the next page.
4 / 12 sections
RESUME ↗
§ YOUR PATH · IMPLEMENTATION
est. 9 weeks remaining
MOD 05
Backprop by Hand
✓ done
→
MOD 06
CNNs from Conv2d up
✓ done
→
MOD 07
Attention & Transformers
34%
→
MOD 08
Diffusion Models
queued
→
§ NOTEBOOKS · RECENT
nb
8 cells
attention.heatmap
edited 12 min ago
nb
5 cells
softmax.stability
edited 2 hr ago
nb
11 cells
kv-cache.memory
edited yesterday
§ PROBLEM QUEUE
4 due
14.2.A
Scaled dot-product attention
med
in progress
→
14.2.B
Implement masked attention
med
→
13.4.C
Layer norm vs batch norm
easy
review
→
14.3.A
Multi-head from scratch
hard
→
§ ACTIVITY · LAST 12 WEEKS
less
more
NEXT MILESTONE
Finish Module 07 — earn the "Wrote a transformer" badge.
5 of 16 chapters
~ 11 days