Where inside the attention sub-layer is dropout typically applied during training, and what does dropping there teach the model? Mention the dominant placement and any secondary placements.
Free · 2 AI evals / day
Attention dropout sits in a very specific place: right after the softmax that produces the attention weight matrix.
The attention output is computed in three steps:
1. Raw scores: `S = Q K^T / sqrt(d_k)`.
2. Weights: `A = softmax(S)`, each row isa probability distribution over keys.
3. Output: `out = A V`.
Dropout is applied to A in step 2 before step 3. Each entry of A is independently set to zero with probability p, and the surviving entries are scaled up by `1 / (1 - p)` so the expected sum stays roughly the same.
What does dropping there teach? It discourages **single-key collapse**. If a query has learned to pour all its weight on one specific key, dropout will sometimes zero that key out and force the query to recover by spreading weight across other keys. Over training this regularizes the attention pattern to be more robust.
There is also a secondary dropout slot after the output projection `W_O`, but that one is just generic residual dropout, not attention-specific. Most modern LLM pretraining (Llama, Mistral, Qwen) actually uses zero attention dropout because the data scale alone is enough regularization.
The dominant attention dropout slot is on the **post-softmax weight matrix** `A = softmax(Q K^T / sqrt(d_k))`. Each entry is zeroed independently with probability p, and the surviving entries are scaled by `1 / (1 - p)` to preserve the expected weighted sum.
The semantic effect is what makes this placement special. Dropping individual attention weights forces each query to **spread its mass** across multiple keys instead of collapsing onto a single sharp peak. If the model has learned a brittle 'always look at token 7' pattern, dropout breaks that pattern often enough that gradient descent has to find a more distributed solution. It is regularization specifically against over-confident, narrow attention.
A secondary slot sits **after the output projection** `W_O`, on the residual-stream activations leaving the sub-layer. That one is generic residual dropout, not attention-specific, and it has the same role here as it does anywhere else in the network.
The two slots use independent dropout rates in most reference implementations. Vaswani's original transformer used 0.1 in both. Modern frontier LLMs (Llama, Mistral, Qwen, DeepSeek) typically set attention dropout to 0 at pretraining scale, the data islarge enough that explicit regularization stops being the bottleneck, but supervised fine-tuning recipes still use it because the data is smaller and overfitting is easier.
Quick reference
TL;DR
Attention dropout is applied to the post-softmax weight matrix to discourage single-key collapse; a second slot lives after the output projection as residual dropout.
Imagine a student doing a group project who only ever asks one teammate for help. When the test comes, that teammate is sick and the student panics. A teacher who randomly tells the student 'pretend that teammate is unavailable today, work with someone else' is forcing the student to learn to ask many teammates. That is what attention dropout does. It randomly silences some of the attention links during training so the model has to spread its bets and not get addicted to one particular key. At test time everyone is available, but the student has learned not to depend on any single source.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example. Click to expand.
Everything you need to truly understand this topic: intuition, mechanics, step by step explanation, code, formulas, and worked example.
Trillion-token pretraining on diverse web data Implicit regularization from data diversity already prevents over-fitting; explicit dropout slows convergence without a measurable quality gain.
Inference time on any model Dropout is a training-only regularizer; at inference all attention weights flow through unchanged. Forgetting to disable it produces stochastic outputs that look like a sampling bug.
Mixture of experts attention layers with routing-based regularization MoE adds its own load-balancing and routing-noise regularization; stacking attention dropout on top often hurts the routing dynamics without adding meaningful generalization.
Quick scan · 2–4 min
Everything important, quickly.
Place attention dropout on the post-softmax weight matrix, explain the spread the mass semantics, note the secondary residual slot after W_O, and close with the modern pretraining trend toward 0.0.
Real-world examples
Real products, models, and research that use this idea.
1Vaswani 2017 original Transformer: dropout 0.1 on the post-softmax weight matrix and 0.1 residual dropout after each sub-layer.
2Llama 4 Maverick, DeepSeek V4, Mistral Large 2 pretraining: attention dropout typically set to 0.0 because data scale provides implicit regularization.
3Instruction fine-tuning recipes (LoRA on Llama 4, SFT on Qwen 3.5): attention dropout often 0.05-0.1 to prevent overfitting on small SFT datasets.
4PyTorch's `nn.MultiheadAttention(dropout=p)` and `F.scaled_dot_product_attention` both apply dropout at the post-softmax weight slot.
Interviewer follow-ups
What an interviewer would ask next. Try answering before peeking at the approach.
QIf you wanted to regularize attention without using dropout, what alternatives exist?
Sign in or go Pro to see the approach for this follow-up.
QWhy might attention dropout hurt at extreme pretraining scale?
Sign in or go Pro to see the approach for this follow-up.
Don't say this
Red flags and common mistakes that signal junior thinking. Click to expand.
Most common mistake
Applying dropout on Q, K, or V before the dot product instead of on the post-softmax weights. The standard attention-specific slot is on A, not on the inputs.