#Transformer
28 related articles
MDT Model: Detecting Ambivalence and Hesitancy Through Cross-Modal Conflict Recognition
MDT (Modality Discrepancy Transformer) explicitly models cross-modal conflicts across face, speech, and language to recognize ambivalence and hesitancy, achieving 0.7408 Macro F1 on BAH — 10+ points above the strongest baseline — in under 20 minutes on a single GPU.
Understanding KV Cache: The Space-Time Tradeoff Behind LLM Inference Optimization
A deep dive into KV Cache for LLM inference: how it eliminates redundant Key/Value recomputation in autoregressive generation, with complexity analysis and memory estimation formulas.
Transformer Attention Mechanism Deep Dive: From Tokens to Self-Attention, Step by Step
A deep dive into the Transformer attention mechanism — from tokenization and embedding to Q/K/V self-attention, Mask, and Softmax — with step-by-step dimension tracking.
Decoding the Mathematical Framework for Transformer Circuits: A Foundational Work in Mechanistic Interpretability
A deep dive into Anthropic's Mathematical Framework for Transformer Circuits — covering QK/OV circuits, residual streams, and induction heads in mechanistic interpretability.
Deep DivesDemystifying Transformer: A Word-Continuation Function, Deconstructed
Understand Transformer through the lens of word continuation. Breaking down language generation into Embedding, Transformer Block, and Probability output modules for intuitive understanding.
Deep DivesThe "Worse is Better" Philosophy of Large Model Design: Why Simple and Brutal Beats Refined and Complex
Analyzing the "worse is better" philosophy in large model architecture: why DeepSeek V4 dropped N-gram, why Transformer dominates AI, and three iron laws of simple, efficient model design.
Deep DivesCore Principles of the Transformer Architecture: A Deep Dive into Self-Attention Mechanisms and Engineering Optimizations
Deep dive into Transformer architecture covering self-attention QKV mechanics, Encoder-Decoder structure, Flash Attention memory optimization, RoPE positional encoding, and GQA inference acceleration.
Deep DivesReAct Pattern and Transformer Architecture Explained: High-Frequency AI Interview Topics for Frontend Developers
Deep dive into Agent core logic ReAct (Reasoning + Acting) execution flow, Transformer self-attention mechanism, and essential AI skills for frontend developers including prompt engineering, RAG, and Agent development.