Name what wraps the attention sub-layer in every transformer block.

Flashcard·Easy·4.0 · 0·~30s·Asked atDeepseek Paytm·Relevant atMicrosoft

Attempt it

Two things wrap attention in every transformer block: a LayerNorm (or RMSNorm) and a residual connection. The standard pre-norm pattern is x + Attention(LN(x)). LN is applied to the input first, then attention runs, then the result is added back to the original x. The residual lets gradients flow cleanly through dozens of layers; without it, deep transformers do not train. Modern decoder-only LLMs like Llama and Mistral use pre-norm with RMSNorm; the original 2017 transformer used post-norm with LayerNorm.

Your answer

Every transformer block has a standard wrapper around the attention sub-layer. It has two pieces: 1. **A normalization layer** (LayerNorm in the original 2017 transformer; RMSNorm in modern Llama, Mistral, Gemma). It standardizes the per token activation vector to a known scale before attention sees it. 2. **A residual connection**, the skip path that adds the original input back to the attention output: `x + Attention(LN(x))`. The modern pattern is **pre-norm**: normalize first, then run attention, then add to the residual. The original 2017 transformer used **post-norm**: run attention first, add residual, then normalize. Pre-norm is more stable for very deep stacks and is the default in every modern LLM. The residual is the key piece. Without it, gradients have to flow through every attention sub-layer in series, and they degrade across the stack. Networks beyond a handful of layers stop training. With the residual, the gradient can take the skip path directly back to early layers. This is what makes 70- to 100-layer transformers trainable at all. The normalization keeps activations in a stable range so attention's softmax does not saturate. That is the other half of why the wrapper is universal.

There are always two wrappers: a normalization and a residual connection. The modern pre-norm pattern is `output = x + Attention(LN(x))`. The input x goes through normalization (LayerNorm or RMSNorm), then attention. Then the skip connection adds the attention output back to the original x. The same wrapper pattern repeats for the FFN sub-layer: `output = x + FFN(LN(x))`. The normalization (`LN`) keeps per token activation vectors at a stable scale before attention reads them. Without it, dot-product magnitudes drift, softmax saturates, and gradients vanish on non-peak keys. LayerNorm centers and scales; RMSNorm drops the centering and the bias, keeping only the divide-by-RMS plus a learnable scale. Llama and Mistral use RMSNorm because it is cheaper with no measurable quality loss. The residual connection is the architectural reason transformers train at depth. Without `x + ...`, the only path for the gradient is back through every attention block in series, and deep transformers do not train. With the residual, the gradient has a clean identity path back to earlier layers (a constant-1 derivative through the skip), so 70-, 80-, 96-layer stacks are stable. Pre-norm vs post-norm matters: the original 2017 transformer used post-norm (`LN(x + Attention(x))`), which works but requires careful learning rate warmup. Pre-norm is the modern default because it allows training without warmup and scales better.

TL;DR

Every transformer block wraps attention in two things: a normalization (LayerNorm or RMSNorm) on the input and a residual skip that adds the attention output back to the original input.

Memory aid

Easy to grasp

Picture a long assembly line where each station does something to a product, but the product is also carried past every station on a conveyor belt. Each station gets to see the product, do its little adjustment, and then add that adjustment to whatever was on the belt. If the station messes up, the original product still arrives at the end. That conveyor belt is the residual stream. And before any station touches the product, an inspector standardizes its size so the station gets predictable input. That is the normalization. Both wrappers exist because without them deep stacks of stations would either break the product completely or fail to train at all.

Key concepts

Residual connection LayerNorm / RMSNorm Transformer block

Concept explanation~2 min read

Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.

Every transformer block wraps its attention sub-layer in two pieces of architectural plumbing: a normalization and a residual connection. These are not optional decorations. They are the reason deep transformers train at all. The small variations across them (pre-norm vs post-norm, LayerNorm vs RMSNorm) explain a surprising fraction of empirical performance differences across model families.

The wrapper formula and what each piece does

The standard modern transformer block, in pseudocode:

python

def transformer_block(x):
    x = x + attention(rmsnorm(x))
    x = x + ffn(rmsnorm(x))
    return x

There are two sub-layers (attention and FFN), each wrapped identically: normalization on the input, sub-layer applied, result added back to the original via the residual skip.

What the normalization does

LayerNorm or RMSNorm standardizes the per token activation vector. Without it, the activation scale can drift across layers (some heads push it up, others pull it down). The dot products inside attention then land in regimes where softmax saturates. The norm keeps the scale predictable so attention sees inputs at a roughly fixed magnitude regardless of layer depth.

What the residual does

The x + ... is the skip connection. The original input passes through unchanged; the sub-layer's contribution is ADDED to it rather than replacing it. The output preserves the input, with the sub-layer's delta layered on top.

This is the same idea as ResNet from 2016 vision: an identity path through the depth of the network that guarantees gradient flow back to early layers.

Why the residual is key

Pre-norm vs post-norm

LayerNorm vs RMSNorm

The residual stream framing

Situations where this technique stops working.

2–4 min · Everything important, quickly.

Real products, models, and research that use this idea.

Llama 4 Maverick uses pre-norm RMSNorm wrapping attention and FFN sub-layers across all 48 blocks.
Mistral 7B ships pre-norm RMSNorm with sliding-window attention inside the wrapper.

What an interviewer would ask next. Try answering before peeking at the approach.

QWalk through how the gradient flows through a residual connection and explain why this prevents vanishing gradients.

For y = x + F(x), the Jacobian dy/dx = I + dF/dx. Stacking blocks gives a product of (I + dF_i/dx) terms. Expanded, this product has a leading I (the identity path). That term ensures the gradient cannot vanish to zero across depth, regardless of what F does.

1 more follow-up an interviewer would ask next. Sign in to reveal them.

Red flags & common mistakes

The phrases that signal junior thinking. Click to expand.

Most common mistake

Forgetting the residual or treating it as 'just nice to have'. Without the residual, gradients vanish across stacked layers and the network does not train.

60 second bullets to scan on the way to the call.

Name both wrappers: normalization and residual
Difference between pre-norm and post-norm

Primary sources. Browse if you want the original framing.