APLAB.ACADEMY · 交互式学习仪器64 TOOLS · 8 COURSES · 97 LESSONS
图 01 — 论点
掌握
基础。
通过交互。
大多数 ML 课程要求你阅读公式并观看视频。这里，每节课都围绕一个可操作的仪器构建——一个你可以一边阅读一边摆弄的东西。先建立直觉，然后是数学，然后是代码。
FIG. 01Multi-Head Self-Attention — try a token
INTERACTIVE
YOUR SENTENCE · TYPE TO POKE
INPUT TOKENS · CLICK TO QUERY
ATTENTION FROM "sat" — HEAD 0
1
99
1
0
0
0
The
cat
sat
on
the
mat
Illustrative — stylised attention patterns computed from your sentence, not a trained model. The real engine is in the Attention Microscope.
ATTENTION HEADS
OBSERVATION · PREVIOUS TOKEN
This head looks one step back — each token attends to the token immediately before it. The very first token has nothing prior, so it attends to itself. Induction and copying circuits lean on exactly this.
Figure 1.来自 Transformer Explorer 的实时预览。点击词元，切换注意力头，观察模式变化。
§02 · 方法
连接式学习，而非线性讲座。
每个概念有四次穿越。每次都让你已经理解的内容更深一层。
I.
直觉
看到它运作。
从可操作的可视化开始。先不上方程——只观察模式。
↳ interact
II.
理论
然后形式化。
当直觉建立后，数学就成为对你已看到的内容的描述。
↳ derive
III.
代码
亲手实现。
在可运行的环境中实现概念。打破它。修复它。掌握它。
↳ execute
IV.
连接
它还在哪里？
每个概念都连接到它在其他地方的出现——是知识图谱，而不是轨道。
↳ navigate
FIG. 01VISUALIZATION
Model Training & Parallelism Explorer
Comprehensive tool for exploring training strategies
使用于 1 节课打开 →
FIG. 02INTERACTIVE
AI Agents Explorer
Watch AI agents move, think, and communicate in real-time
使用于 1 节课打开 →
FIG. 03INTERACTIVE
Python Memory Explorer
Deep dive into Python memory management, object model, and CPython internals
使用于 1 节课打开 →
FIG. 04INTERACTIVE
Tokenizer Bench
Same text, four tokenizers. Token chunks, costs, and a step-by-step BPE merge view.
使用于 1 节课打开 →
FIG. 05VISUALIZATION
Gradient Lab
2D classification with a hand-rolled MLP. Step through SGD, watch the decision boundary form.
使用于 2 节课打开 →
FIG. 06VISUALIZATION
Attention Microscope
Inspect synthetic GPT-2-style attention layer by layer, head by head. Ablate heads, watch top-5 candidates shift.
使用于 2 节课打开 →
§04 · 课程
八门课程。从基础到前沿。
结构化路径——但每个概念也是一个节点，可从任何地方访问。
№课程分类课时级别
01Python Fundamentals: Programming FoundationsPython 编程·12 lessons·初级12初级→02Python for Data Science: From Arrays to AnalysisPython 编程·10 lessons·中级10中级→03Python Advanced: Professional Engineering MasteryPython 编程·12 lessons·高级12高级→04Classical Machine Learning: Supervised Learning Foundations机器学习·15 lessons·中级15中级→05Advanced ML: Unsupervised Learning & Production机器学习·12 lessons·高级12高级→06NLP Fundamentals: Core Concepts and Architectures自然语言处理·11 lessons·中级11中级→07Advanced NLP: Training & Production Systems★ 高级自然语言处理·11 lessons·高级11高级→08AI Agents: Building Autonomous Intelligent SystemsAI 智能代理·14 lessons·高级14高级→