NLP Academy · 课时

Transformer 模块内部

了解完整架构如何组合起来

Transformer 模块内部 是 CoddyKit 上的免费 NLP Academy 课时。这是第 4 节课，共 4 节。你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。这是 NLP Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 NLP Academy 课程共包含 4 节课。

本课时的部分内容尚未翻译，以英文显示。

Stacking Blocks

A Transformer is built by stacking the same block many times. Each block refines the representation a little more. 🧱

Two Main Sublayers

Every block has two parts: a multi-head attention sublayer followed by a small feed-forward network applied to each position.

The Feed-Forward Network

The feed-forward sublayer is two linear layers with a nonlinearity between them, processing each token independently for extra power.

h = relu(x @ W1 + b1) @ W2 + b2

Residual Connections

Each sublayer wraps its input in a residual connection: add the input back to the output so gradients flow and nothing is lost.

out = x + sublayer(x)

Layer Normalization

After adding the residual, layer normalization rescales the values. This keeps activations stable and speeds up training.

out = layer_norm(x + sublayer(x))

Encoder Blocks

An encoder stack reads the input and builds rich context vectors. It uses self-attention so every token sees all the others.

Decoder Blocks

A decoder generates output one token at a time. It adds masked attention plus cross-attention back to the encoder.

Masked Attention

During generation, masking hides future tokens so the decoder cannot peek ahead and must predict the next word honestly.

Cross-Attention

Cross-attention lets the decoder query the encoder's outputs, connecting what it is writing to what it originally read.

Depth Brings Power

Stacking many blocks lets early layers catch simple patterns and later layers build abstract meaning, much like deep vision networks.

Encoder, Decoder, or Both

BERT uses only the encoder, GPT uses only the decoder, and translation models use both. The block is the shared building unit. 🚀

Quick Check

Let us check the block structure.

Recap

You assembled the Transformer block: attention plus feed-forward, wrapped in residuals and layer norm, stacked into encoders and decoders. ✨

常见问题解答

「Transformer 模块内部」课时是免费的吗？

是的 — 「Transformer 模块内部」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 NLP Academy 课程的其余内容，请升级到 CoddyKit PRO。 NLP Academy 课程共包含 4 节课。

「Transformer 模块内部」这节课中我会学到什么？

了解完整架构如何组合起来你通过在浏览器中直接运行的动手代码来练习 NLP Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。

学习 NLP Academy 需要有经验吗？

无需任何先前经验。CoddyKit 上的 NLP Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。这是第 4 节课，共 4 节。

「Transformer 模块内部」课时需要多长时间？

大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。

我能在这节 NLP Academy 课中编写并运行代码吗？

能。每节 NLP Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。

此课程中的所有课时

← 返回 NLP Academy