Skip to content
Tagged articles

LLM Architecture

19 articles · Page 1 of 1
IT Services Circle
IT Services Circle
Nov 6, 2026 · Artificial Intelligence

Beyond Caching: Architectural Solutions for Agent Context Compression Losing Tool Calls

The article analyzes why context compression in AI agents loses tool call history, causing duplicate executions, and presents five architectural fixes: an independent action ledger, idempotent interception with signatures, explicit compression prompts, external storage with retrieval, and hierarchical summarization.

AI agentsAction LedgerContext Management
0 likes · 10 min read
Beyond Caching: Architectural Solutions for Agent Context Compression Losing Tool Calls
Tech Freedom Circle
Tech Freedom Circle
Oct 30, 2026 · Artificial Intelligence

Ant Group RAG Interview: Vector, GraphRAG, LLM-Wiki, PageIndex & Ontology Fusion

This article details five RAG paradigms — Vector RAG, GraphRAG, PageIndex, LLM-Wiki, and Ontology — their core differences, failure modes, and fusion strategies using a time-space orthogonal framework, with code examples and scenario-based orchestration for enterprise-grade retrieval systems.

GraphRAGKnowledge RetrievalLLM Architecture
0 likes · 29 min read
Ant Group RAG Interview: Vector, GraphRAG, LLM-Wiki, PageIndex & Ontology Fusion
SmartApe AI
SmartApe AI
Oct 28, 2026 · Artificial Intelligence

Qwen 4 Imminent: Leaked Code & Anonymous Models Reveal Architecture Details

Community clues suggest Alibaba's Qwen 4 is near release, with inference optimizations in vLLM and llama.cpp, architectural details like a 27B dense model with 5120 hidden dim, 24 query heads, 4 KV heads, 4 residual streams, MTP, and QSA sparse attention, while anonymous Arena models Kalmia and Kerria are speculated as Qwen 4 variants showing strong code and 3D generation.

AlibabaArena benchmarkFP8 quantization
0 likes · 7 min read
Qwen 4 Imminent: Leaked Code & Anonymous Models Reveal Architecture Details
JD Tech
JD Tech
Oct 20, 2026 · Artificial Intelligence

Jev: When AI Makes Decisions Instead of Writing Answers

This article analyzes TypeSafe AI's Jev, a non-autoregressive 'System One' model that returns typed decisions with probabilities instead of generating text, contrasting it with LLMs, detailing its three output primitives (Choice, Score, Noul), four pilot scenarios, a four-step production integration process, and implications for AI software architecture.

AI EngineeringAmdahl's LawDecision Models
0 likes · 29 min read
Jev: When AI Makes Decisions Instead of Writing Answers
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Oct 11, 2026 · Artificial Intelligence

DeepSeek's Cost Curve Chronicle: How Token Economics Rewrote LLM Architecture

This article chronicles DeepSeek's paper timeline from 2023 to 2026, analyzing how each model generation optimized training compute, KV cache, and throughput via MoE, MLA, sparse attention, and conditional memory, shifting long-context competition from maximum length to per-token byte cost.

DeepSeekGRPOKV cache optimization
0 likes · 37 min read
DeepSeek's Cost Curve Chronicle: How Token Economics Rewrote LLM Architecture
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Oct 9, 2026 · Artificial Intelligence

DeepSeek's Accuracy Varies by Token Position: Phase Sensitivity in KV Cache Compression

ByteDance's Seed team discovered that DeepSeek-V4 models exhibit periodic accuracy fluctuations every 4 tokens in long-context retrieval, caused by chunked KV cache compression creating phase sensitivity where retrieval performance depends on information position relative to compression window boundaries, a phenomenon mitigated but not eliminated by post-training.

ByteDance SeedKV Cache compressionLLM Architecture
0 likes · 12 min read
DeepSeek's Accuracy Varies by Token Position: Phase Sensitivity in KV Cache Compression
Machine Heart
Machine Heart
Sep 8, 2026 · Artificial Intelligence

SMELT: Looped Transformers Outperform Baselines Under Fair Budget Matching

The SMELT framework from Tsinghua and ByteDance Seed fairly compares Looped Transformers against baselines by matching compute, parameters, and KV cache, finding that looping the middle 50% of layers twice with a larger depth-width ratio consistently reduces validation loss across scales, saves 6.8–18% training compute, and yields downstream gains beyond loss reduction.

Budget MatchingLLM ArchitectureLooped Transformer
0 likes · 12 min read
SMELT: Looped Transformers Outperform Baselines Under Fair Budget Matching
Chengwu Tech Stack
Chengwu Tech Stack
Sep 1, 2026 · Artificial Intelligence

Agent vs Prompt+LLM: Where Should Uncertainty Live in Production LLM Systems?

This article analyzes when to use autonomous agents versus prompt-engineered LLM pipelines in production systems, arguing that the choice depends on uncertainty type, failure cost, reversibility, and audit requirements, and presents a layered architecture where code handles deterministic boundaries while LLMs handle semantic judgments.

AI system designAgent vs Prompt engineeringLLM Architecture
0 likes · 25 min read
Agent vs Prompt+LLM: Where Should Uncertainty Live in Production LLM Systems?
Geek Labs
Geek Labs
Apr 26, 2026 · Artificial Intelligence

Three Cutting‑Edge Open‑Source Projects Redefining AI Infrastructure

The article reviews three advanced open‑source projects—LingBot‑Map for real‑time 3D scene reconstruction, Browser‑Harness enabling AI‑written browser tools, and OpenMythos recreating Claude Mythos’s looped transformer—showing how AI is shifting toward task execution, 3‑D perception, and deeper architectural innovation.

AI agent automationClaude MythosLLM Architecture
0 likes · 11 min read
Three Cutting‑Edge Open‑Source Projects Redefining AI Infrastructure
Machine Learning Algorithms & Natural Language Processing
Machine Learning Algorithms & Natural Language Processing
Apr 8, 2026 · Artificial Intelligence

Dissecting Gemma‑4’s Architecture and Training Choices: A Technical Comparison with Qwen‑3 and GLM‑5

This article breaks down every architectural and training decision behind Gemma‑4—KV sharing, p‑RoPE, per‑layer embeddings, and a dual‑path MoE + dense MLP—while contrasting its efficiency and performance with Qwen‑3 and GLM‑5 across benchmarks, quantization strategies, and RL pipelines.

GLM-5Gemma 4LLM Architecture
0 likes · 23 min read
Dissecting Gemma‑4’s Architecture and Training Choices: A Technical Comparison with Qwen‑3 and GLM‑5
DeepHub IMBA
DeepHub IMBA
Mar 27, 2026 · Artificial Intelligence

AI Agent Architecture: Chain‑of‑Thought, ReAct, and Tool Calls

From a simple black‑box view where an agent receives a user request and returns an answer, the article breaks down modern AI agent designs—detailing the pure Chain‑of‑Thought reasoning loop, the ReAct reasoning‑acting cycle, tool integration, iteration tuning, and how to choose the optimal architecture for production.

AI agentsLLM ArchitectureReact
0 likes · 9 min read
AI Agent Architecture: Chain‑of‑Thought, ReAct, and Tool Calls
Alibaba Cloud Developer
Alibaba Cloud Developer
Mar 23, 2026 · Artificial Intelligence

From Scenario Abstraction to an AI Assistant Production Line: Scalable Architecture and Prompt Plug‑In Design

The article analyzes the inefficiencies of building isolated AI assistants for each business need, abstracts four high‑frequency scenarios, proposes a reusable technical solution stack—including IntentResult modeling, FSWW tool‑recall, ReAct reasoning, multimodal RAG, and a prompt plug‑in framework—and demonstrates how a one‑click platform can turn these designs into production‑ready AI assistants.

AI assistantsKnowledge RetrievalLLM Architecture
0 likes · 21 min read
From Scenario Abstraction to an AI Assistant Production Line: Scalable Architecture and Prompt Plug‑In Design
DeepNoMind
DeepNoMind
Feb 7, 2026 · Artificial Intelligence

Decoding Claude Skills: From Prompt Engineering to Context Engineering

This article analyzes Anthropic's Claude Skills architecture, showing how moving from monolithic prompt engineering to modular context engineering reduces token waste, improves security, and enables version‑controlled, composable AI agents through progressive disclosure, sandbox isolation, and a three‑tier ledger model.

Claude SkillsLLM ArchitectureModular AI
0 likes · 21 min read
Decoding Claude Skills: From Prompt Engineering to Context Engineering
Code Archaeology
Code Archaeology
Oct 24, 2025 · Artificial Intelligence

DeepSeek's Three Pillars: MoE, MLA, and Multi-Token Prediction Demystified

This article explains DeepSeek's three core innovations—DeepSeekMoE with fine-grained experts and shared experts, Multi-Head Latent Attention (MLA) for compressed KV-cache, and Multi-Token Prediction (MTP) for faster training convergence—detailing how each reduces compute costs and memory overhead while maintaining model performance.

DeepSeekKV cache optimizationLLM Architecture
0 likes · 5 min read
DeepSeek's Three Pillars: MoE, MLA, and Multi-Token Prediction Demystified
Code Archaeology
Code Archaeology
Oct 20, 2025 · Artificial Intelligence

LLaMA 2 & 3 Deep Dive: Architecture Innovations (RMSNorm, RoPE, GQA) and Post-Training Pipeline

This article explains Meta's LLaMA model series, detailing LLaMA 2's architectural innovations—RMSNorm, SwiGLU, rotary positional encoding, and grouped-query attention—and LLaMA 3's post-training pipeline including supervised fine-tuning, reward modeling, rejection sampling, and direct preference optimization.

DPOGrouped-Query AttentionLLM Architecture
0 likes · 10 min read
LLaMA 2 & 3 Deep Dive: Architecture Innovations (RMSNorm, RoPE, GQA) and Post-Training Pipeline
Data Thinking Notes
Data Thinking Notes
Aug 26, 2025 · Artificial Intelligence

From Prompt to Context: How AI Agents Evolve into Proactive Intelligence

This article explores the rapid growth of large language models and explains how AI agents transform passive, single‑turn responses into proactive, continuous intelligence by leveraging a core “Prompt→Context→Action” loop, detailing their architecture, key components, challenges, and future directions.

AI AgentContext ManagementLLM Architecture
0 likes · 20 min read
From Prompt to Context: How AI Agents Evolve into Proactive Intelligence
Volcano Engine Developer Services
Volcano Engine Developer Services
Aug 26, 2025 · Artificial Intelligence

From Single LLM to Multi‑Agent: How Context Engineering Drives the Next AI Architecture

This article examines the evolution of LangChain's Open Deep Research project from a monolithic LLM pipeline to a multi‑agent system, highlighting the role of context engineering, architectural trade‑offs, practical code examples, and best‑practice guidelines for building scalable, token‑efficient AI solutions.

AI researchLLM ArchitectureLangChain
0 likes · 16 min read
From Single LLM to Multi‑Agent: How Context Engineering Drives the Next AI Architecture
Software Engineering 3.0 Era
Software Engineering 3.0 Era
Jul 28, 2025 · Artificial Intelligence

How Enterprise AI Agents Move From Efficiency Silos to Value Resonance

The article explains how enterprise AI agents shift from a command‑response model to a goal‑execution paradigm, outlines a layered LLM architecture, tool integration, and memory systems, and demonstrates three practical use cases that create clear value loops for R&D, marketing, and customer service.

AI agentsLLM ArchitectureMemory Systems
0 likes · 13 min read
How Enterprise AI Agents Move From Efficiency Silos to Value Resonance
DaTaobao Tech
DaTaobao Tech
Sep 11, 2023 · Artificial Intelligence

Large Language Model Upgrade Paths and Architecture Selection

This article analyzes upgrade paths of major LLMs—ChatGLM, LLaMA, Baichuan—detailing performance, context length, and architectural changes, then examines essential capabilities, data cleaning, tokenizer and attention design, and offers practical guidance for balanced scaling and efficient model construction.

BaichuanChatGLMLLM Architecture
0 likes · 32 min read
Large Language Model Upgrade Paths and Architecture Selection