Tag: #ai-papers
Writing on GPUs, LLMs, MLOps, Kubernetes — and mindset · 79 posts
From RLHF to DPO: A Deep Dive into LLM Alignment Techniques
A comprehensive survey of key LLM alignment papers. We analyze the InstructGPT RLHF pipeline, Anthropic Constitutional AI, the mathematical foundations of DPO, PPO training stability, and recent methods like KTO, IPO, an
2026-03-13 · 12 min read #ai-papers#rlhf#dpo#alignment#ppoRAG Paper Survey: The Evolution of Retrieval-Augmented Generation — From RETRO to Self-RAG and Corrective-RAG
Traces the evolution of RAG research through papers. Compares architectures and benchmarks from the original RAG to RETRO large-scale retrieval, Self-RAG self-reflection, and Corrective-RAG retrieval quality evaluation.
2026-03-12 · 22 min read #ai-papers#rag#self-rag#corrective-rag#retroDiffusion Model Paper Survey: Evolution of Image Generation from DDPM to Stable Diffusion, DiT, and SDXL
A comprehensive survey covering the DDPM/DDIM forward-reverse diffusion theory, score-based models, Latent Diffusion (Stable Diffusion) VAE+U-Net architecture, Classifier-free Guidance, DiT (Diffusion Transformer) adaLN-
2026-03-12 · 18 min read #ai-papers#diffusion-model#ddpm#stable-diffusion#ditMixture of Experts (MoE) Architecture Deep Analysis: Evolution from Switch Transformer to Mixtral and Efficient Scaling Strategies
Deep analysis from the core principles of MoE architecture to Switch Transformer single expert routing, Mixtral 8x7B Sparse MoE, and DeepSeek-MoE fine-grained strategy. Covers routing mechanisms, load balancing loss, tra
2026-03-11 · 17 min read #ai-papers#moe#switch-transformer#mixtral#model-architectureKV Cache Optimization Deep Dive: GQA, MLA, and MHA Attention Mechanisms with Memory Efficiency Strategies
From Transformer Self-Attention KV Cache fundamentals through MHA, MQA, GQA (Llama 2/3), and MLA (DeepSeek-V2/V3) mechanism memory analysis and comparison, KV Cache compression techniques (quantization, eviction policies
2026-03-11 · 17 min read #ai-papers#kv-cache#attention-mechanism#gqa#mlaDeep Dive into Mixture of Experts (MoE) Architecture: From Switch Transformer to Mixtral and DeepSeek
A comprehensive analysis of Mixture of Experts (MoE) architecture. Covers the mathematical foundations of Sparse MoE, routing strategies in Switch Transformer, Mixtral 8x7B, and DeepSeek-V3, training stability techniques
2026-03-10 · 15 min read #ai-papers#moe#transformer#mixtral#deepseekFlashAttention Paper Analysis: Revolutionizing Transformer Training and Inference with IO-Aware Exact Attention
An in-depth analysis of the FlashAttention series (v1 through v3). Covers the tiling strategy of the IO-aware algorithm, GPU SRAM/HBM memory hierarchy utilization, backward pass recomputation, parallelism improvements in
2026-03-09 · 22 min read #ai-papers#flash-attention#transformer#gpu-optimization#attention-mechanismDPO (Direct Preference Optimization) Paper Deep Analysis — LLM Alignment Without RLHF
From mathematical principles of DPO to implementation, comparison with RLHF, and variants like IPO/KTO/ORPO — an in-depth analysis of LLM preference optimization from a practical perspective.
2026-03-09 · 20 min read #ai-papers#dpo#rlhf#llm-alignment#preference-optimizationMath + LaTeX/KaTeX Complete Guide for Reading AI/ML Papers
A comprehensive summary of essential math concepts (linear algebra, calculus, probability/statistics, optimization) and LaTeX/KaTeX formula syntax for reading AI/ML papers with practical examples, symbol cheat sheets, an
2026-03-08 · 20 min read #ai-papers#math#latex#katex#linear-algebraKAN (Kolmogorov-Arnold Networks) Paper Analysis: Learnable Activation Function Architecture Replacing MLPs
In-depth analysis of the KAN paper. Covers the mathematical background of the Kolmogorov-Arnold representation theorem, B-spline activation functions, accuracy and interpretability comparison with MLPs, and practical imp
2026-03-08 · 38 min read #ai-papers#kan#kolmogorov-arnold#neural-network#mlpRing Attention Paper Analysis: Implementing Infinite Context Window Training in Distributed Environments
Analyzes the Ring Attention paper exploring methods to overcome context length limitations in distributed environments. Covers the connection with Blockwise Parallel Transformer, implementation details, performance bench
2026-03-08 · 33 min read #ai-papers#ring-attention#distributed-training#long-context#transformerMamba and State Space Model Paper Deep Dive: Transformer Alternative Architectures from Selective SSM to Mamba-2
A comprehensive analysis of the SSM architecture covering the optional State Space Model mechanism of the Mamba paper, the development process from S4 to Mamba-2, the pros and cons of linear time complexity compared to T
2026-03-07 · 30 min read #ai-papers#mamba#state-space-model#ssm#transformerRWKV Architecture Deep Dive: Linear Attention RNN That Rivals Transformers
A comprehensive analysis of RWKV architecture covering the WKV attention mechanism, linear complexity advantages, comparison with Transformers and Mamba, training methodology, inference optimization, and practical deploy
2026-03-07 · 19 min read #ai-papers#rwkv#linear-attention#rnn#transformerDeep Dive into Sparse Mixture of Experts (MoE) Architecture: From Design Principles to DeepSeek-V3 and Qwen3
Analyzing the mathematical principles, routing strategies, and load balancing techniques of Sparse MoE architecture, covering the design choices and practical training/inference optimization of modern MoE models from Swi
2026-03-06 · 22 min read #ai-papers#moe#sparse-model#deepseek#2026-03BitNet Paper Analysis: The Era of 1-Bit LLMs — From Ternary Weights to CPU Inference
A comprehensive guide analyzing Microsoft Research's BitNet series (v1, b1.58, a4.8, 2B4T), covering ternary weight training principles, the bitnet.cpp inference framework, and real-world benchmarks.
2026-03-06 · 23 min read #ai-papers#bitnet#1-bit-llm#quantization#model-efficiencyFrom DPO to KTO: Latest Human Feedback Alignment Techniques Paper Review and Practical Implementation
Paper review and TRL-based practical implementation guide covering the latest human feedback alignment techniques such as DPO, IPO, and KTO that overcome RLHF limitations. Algorithm comparison, hyperparameter tuning, and
2026-03-05 · 22 min read #ai-papers#dpo#kto#alignment#2026-03Mamba vs Transformer Practical Comparison: From Paper to Product
Mamba vs Transformer Practical Comparison: From Paper to Product - A comprehensive practical document covering Why/How/When, comparison tables, troubleshooting, hands-on code, and quizzes all in one place.
2026-03-04 · 8 min read #ai-papers#practical-guide#productionAI Paper Reading: Agentic Reasoning Implementation Guide 2026
AI Paper Reading: A practical guide on Agentic Reasoning Implementation Guide 2026 covering Why, How, When, comparison tables, troubleshooting, code examples, and quizzes.
2026-03-04 · 16 min read #ai-papers#2026-03Mixture of Experts (MoE) Architecture Paper Review and Production Scaling 2026
Core paper analysis of Mixture of Experts (MoE) architecture. Covering Switch Transformer, GShard, ST-MoE, and Mixtral -- routing mechanisms, expert parallelism, and load balancing strategies from a production perspectiv
2026-03-04 · 20 min read #ai-papers#2026-03AI Papers: Test-Time Scaling Key Paper Summary
AI Papers: Test-Time Scaling Key Paper Summary - Practical Application Guide as of 2026
2026-03-04 · 24 min read #ai-papers#test-time-scaling#2026-03