The Residual Connection Is Broken. Here's the Fix.
Attention Residuals replaces the standard fixed residual accumulation with softmax attention over previous layer outputs. This enables each layer to selectively combine earlier representations using learned, input-dependent weights. Attention Residuals replaces standard fixed residual accumulation with depth-wise softmax attention over all preceding layer outputs. This enables each layer to combine earlier representations using learned, input-dependent weights. 00:00 Intro to residual connections 03:27 Intuition behind attention residuals 04:43 Full attention residuals 09:43 Block attention residuals 13:07 Parallelism 14:21 Infrastructure design for efficient training 20:03 Infrastructure design for efficient inference 22:01 Discussions 21:02 Related work References: [Attention Residual] https://arxiv.org/abs/2603.15031

The Most Underrated Layer Inside Every AI Model

Spectral Graph Theory For Dummies

【DL輪読会 #365 1/2】Flow Matching for Generative Modeling

Beyond Softmax: The Future of Attention Mechanisms

Advanced Rag Graphrag

How mHC Reinvents Residual Connections
![How LLMs Learn to Reason [GRPO]](https://i.ytimg.com/vi/mg-iU-WxiNs/hqdefault.jpg?sqp=-oaymwEjCNACELwBSFryq4qpAxUIARUAAAAAGAElAADIQj0AgKJDeAE=&rs=AOn4CLA841kCTvje_hnY0iYJP3lGECv2Eg)
How LLMs Learn to Reason [GRPO]

LLMs Don't Need More Parameters. They Need Loops.

We’ve Been Doing Attention Wrong (2-Line Fix)
![How Attention Got So Efficient [GQA/MLA/DSA]](https://i.ytimg.com/vi/Y-o545eYjXM/hqdefault.jpg?sqp=-oaymwEjCNACELwBSFryq4qpAxUIARUAAAAAGAElAADIQj0AgKJDeAE=&rs=AOn4CLBuOQf8Rw0rEDbSy5MucgJ2Vh6xGw)
How Attention Got So Efficient [GQA/MLA/DSA]

The 60-Year Hunt for AI's Most Important Function

Mixture of Experts (MoE), Visually Explained
![This Simple Optimizer Is Revolutionizing How We Train AI [Muon]](https://i.ytimg.com/vi/bO5nvE289ec/hqdefault.jpg?sqp=-oaymwEjCNACELwBSFryq4qpAxUIARUAAAAAGAElAADIQj0AgKJDeAE=&rs=AOn4CLAzFxNYWuTGV6zIBHgFHXfRMkBUNg)
This Simple Optimizer Is Revolutionizing How We Train AI [Muon]

DeepSeek Gave LLMs a Real Memory (It's Not RAG)

How FlashAttention Accelerates Generative AI Revolution

But What Are Transformers?

They solved AI’s memory problem!

DeepSeek V4's Secret: 98% Less Memory
![How Rotary Position Embedding Supercharges Modern LLMs [RoPE]](https://i.ytimg.com/vi/SMBkImDWOyQ/hqdefault.jpg?sqp=-oaymwEjCNACELwBSFryq4qpAxUIARUAAAAAGAElAADIQj0AgKJDeAE=&rs=AOn4CLB6gWS_ZRO-UhithwlfNKgGNDFVNQ)
How Rotary Position Embedding Supercharges Modern LLMs [RoPE]

