Tagged articles
model architecture
85 articles · Page 1 of 1
Tongyi's Qwen4-Max reduces per-task cost from $4.43 to $1.67 (62% less) on its Work360 benchmark while raising scores from 47.74 to 52.69, using a 2.4T-parameter architecture with 710M activated parameters, phrase embeddings, gated recurrent layers, mixture-of-experts, and sparse attention, with training loss declining steadily without spikes.
Qwen4-Maxbenchmarkcompression rate
0 likes · 8 min read

French AI company Mistral releases Mistral Large 4, a 1.05 trillion parameter open-source MoE model with 128 experts and 180B active parameters, achieving 91.2 MMLU-Pro and 89.7% HumanEval+, detailing architecture, training on 16K H200 GPUs, inference via expert offloading, and Apache 2.0 licensing enabling sovereign AI deployment.
1.05T parametersAI sovereigntyMistral Large 4
0 likes · 13 min read

Two senior undergraduates won the Scaling Law Original Breakthrough Award and 10th place in the academic track at the 2026 Tencent Advertising Algorithm Competition by running nearly 1000 experiments, discovering that increasing model width alone reduces AUC unless accompanied by new informative tokens, leading to their CoSFormer architecture that scales capacity and information jointly.
AUCCoSFormerScaling Laws
0 likes · 8 min read


IT XianyuSep 3, 2026 · Artificial Intelligence
Analysis of OpenAI's recurrent depth architecture in Astra shows 3.5B models matching 50B performance by cycling data through reused layers, but raises safety concerns as hidden internal reasoning makes chain-of-thought unauditable, evidenced by a July rogue agent incident detected only through visible reasoning traces.
AI AlignmentAI safetyLOTUS
0 likes · 8 min read

The article reviews DeepSeek‑V4‑Flash‑Vision‑Exp, exposing a misidentification bug, detailing its visual‑primitive approach, impressive spatial‑reasoning benchmarks, and a highly compressed KV‑cache architecture that balances performance with efficiency.
DeepSeekKV Cache compressionMoE
0 likes · 4 min read

The TAAC × KDD Cup 2026 workshop in Jeju Island gathered researchers and engineers from academia and leading tech firms to dissect real‑world recommendation challenges, present the competition’s unified sequence‑modeling and feature‑interaction task, and showcase winning solutions that balance effectiveness, efficiency, and scalability.
KDD CupScaling LawTAAC 2026
0 likes · 7 min read

The article examines the limits of Transformer models, outlines their three fundamental shortcomings, and evaluates the Mobius architecture that decouples knowledge and reasoning, showing up to four‑fold inference speed gains, comparable accuracy, and improved compositional generalisation.
AI EfficiencyContinual LearningMobius
0 likes · 14 min read

The article examines the inherent limitations of the Transformer architecture—its three core shortcomings—and presents the Mobius framework, which decouples knowledge and reasoning to address these issues, offering up to four‑fold inference speed gains while maintaining accuracy and improving data efficiency.
AI scalingMobiusTransformer
0 likes · 14 min read

The article breaks down large‑scale neural‑network training into three error components—data, optimization, and architecture—derives a heterogeneous power inequality, and uses it to analyze optimal learning rates, batch sizes, model size, sparsity, and data scaling, linking the results to known laws such as Kaplan, Chinchilla, Step, Microsoft, and DeepSeek.
Data EfficiencyScaling Lawsdeep learning
0 likes · 28 min read

Shanghai AI Lab’s Shushu team launches ArchSpace, an open‑source platform that releases full model checkpoints, training logs, architecture decisions and validation results, inviting the community to submit and verify foundation‑model innovations such as Inter‑layer Attention and Sparse Attention across 1B‑8B scale models, with transparent logs on W&B and HuggingFace.
AI researchArchSpaceCommunity Validation
0 likes · 10 min read

This article maps the entire large‑model production chain—from data collection, token prediction, and architecture design through training, alignment, inference, multimodal perception, agentic action, deployment, evaluation, and safety—highlighting key engineering decisions, trade‑offs, and concrete examples.
Agent SafetyInference OptimizationRetrieval-Augmented Generation
0 likes · 47 min read

Kimi K3 arrives with 2.8 T parameters, native visual understanding, a 1 M‑token context window, novel KDA and Attention Residuals architecture, aggressive MoE sparsity, and pricing far below Western rivals, while achieving SOTA results on programming and agent benchmarks and demonstrating real‑world research and 3D‑game capabilities.
2.8T parametersAI benchmarksKimi K3
0 likes · 16 min read

The LLM Architecture Gallery, created by Sebastian Raschka, consolidates metadata and standardized visual cards for major large‑language models—from GPT‑2 to trillion‑parameter systems—highlighting architecture trends such as sparse‑mixture‑of‑experts, evolving attention mechanisms, and lightweight alternatives, enabling researchers and developers to compare designs, parameters, licenses, and inference costs in one platform.
Attention MechanismLLMmodel architecture
0 likes · 4 min read

OpenAI’s Images 2.0 (gpt-image-2) replaces the traditional image‑generator model with an interactive creative engine that plans, searches the web, and self‑verifies before rendering, offering higher‑quality multi‑language text, batch consistency, and real‑time information at the cost of a token‑based pricing model and limited access to its most advanced features.
AI image generationCompetitive AnalysisGPT Image 2
0 likes · 32 min read

The article surveys recent open‑weight LLM releases—Gemma 4, Laguna XS.2, ZAYA1‑8B and DeepSeek V4—detailing how KV‑cache sharing, per‑layer embeddings, layer‑wise attention budgeting, compressed convolutional attention and manifold‑constrained hyper‑connections dramatically reduce memory and compute for ultra‑long contexts while preserving model quality.
Attention optimizationKV CacheLLM
0 likes · 25 min read

SenseNova U1 introduces the NEO‑Unify native unified architecture that eliminates separate vision encoders and VAEs, enabling simultaneous multimodal understanding, reasoning, and generation, and achieves state‑of‑the‑art benchmark scores that surpass larger proprietary models across vision‑language, reasoning, and generation tasks.
NEO-unifySenseNova U1benchmark
0 likes · 19 min read

From early 2024 explorations to a 2026 research surge, this review explains how large‑model latent space replaces explicit token‑based processing, outlines its five analytical lenses—foundation, evolution, mechanism, ability, outlook—compares representational properties, details architectural and computational strategies, enumerates new capabilities, and discusses remaining challenges and future directions.
Representation Learningartificial intelligencelarge models
0 likes · 20 min read


SuanNiMar 18, 2026 · Artificial Intelligence
The LLM Architecture Gallery, created by Sebastian Raschka, offers an interactive visual compendium of open‑weight large language models from 2019 to 2026, detailing their core parameters, architectural innovations, and the broader trends shaping modern AI research.
AILLMartificial intelligence
0 likes · 8 min read

This article analyzes the newly released Attention Residuals paper, explaining how learnable attention weighting replaces fixed residual addition to mitigate information dilution in deep LLMs, detailing the proposed Block AttnRes design, engineering trade‑offs, experimental results, and its significance for foundational model architecture.
AttentionBlock AttentionLLM
0 likes · 9 min read

SenseTime eliminates the intermediate encoder in multimodal AI models, allowing direct cross‑modal learning, which yields markedly higher performance at 2‑trillion‑parameter scale while reducing training cost, and may trigger a broader industry move toward simpler, more efficient architectures.
AI paradigm shiftEfficiencySenseTime
0 likes · 6 min read

An AI startup claims that training with only a thousand hours of data produced emergent intelligence and outperformed industry leaders in benchmark tests, prompting a debate over whether this represents a paradigm shift in efficient learning or an overhyped breakthrough requiring further validation.
AIData Efficiencybenchmark
0 likes · 5 min read

MiniCPM‑SALA introduces a hybrid sparse‑linear attention architecture that reduces quadratic compute and memory costs, achieves state‑of‑the‑art performance on long‑context benchmarks, and delivers up to 3.5× faster inference than full‑attention models on sequences up to 1 million tokens.
LLMLinear attentionLong Context
0 likes · 17 min read

This article chronicles the rapid progression of GPT models from the 2018 GPT‑1 pre‑training breakthrough through GPT‑2’s multitask learning, GPT‑3’s scaling laws and few‑shot abilities, to GPT‑4’s multimodal capabilities and the 2024 GPT‑4 Turbo, Sora, and GPT‑4o releases, while also explaining core LLM abilities and the decoder‑only architecture of GPT‑2.
AI evolutionFew-Shot LearningGPT
0 likes · 20 min read

OpenMOSS and MOSI unveiled MOVA, China’s first high‑performance open‑source audio‑video generation model, detailing its dual‑tower architecture, bridge module, aligned ROPE, multi‑stage data pipeline, training strategies, dual CFG guidance, and benchmark results that surpass leading closed‑source systems.
MOVAaudio-video generationmodel architecture
0 likes · 20 min read

This article maps the rapid evolution of open‑source large language models in 2025, explains the underlying architectural breakthroughs such as MLA, MoE, and NSA, compares dozens of models—including DeepSeek‑V3, OLMo2, Gemma3, Llama4, Qwen3, and Kimi‑K2—and highlights the emergence of powerful AI assistants like Dola, providing developers with a concise technical roadmap.
AI assistantLLM efficiencylarge language models
0 likes · 44 min read

The author details Kimi Linear's architecture, training challenges, aggressive MoE sparsity, hybrid linear attention design, benchmark gains, and post‑training insights, offering a transparent technical review of this 48B‑parameter MoE LLM built on 5.7 T tokens.
Hybrid ModelKimi LinearLLM
0 likes · 9 min read

In a Silicon Valley tech conference, Mu Shen shared how voice agents—real‑time, task‑oriented AI—were applied to an open‑world game as an AI NPC and to a Fortune‑500 insurer as an AI tele‑salesperson, revealing technical challenges, model architectures, training strategies, evaluation methods, and key lessons for future deployments.
game AIinsurance automationlarge language models
0 likes · 19 min read

This article revisits nanochat's core components, detailing the preparation of diverse training datasets, the scaling calculations for tokens and parameters, the model's MQA and KV‑cache design, the full training pipeline with gradient accumulation and mixed‑precision, cost breakdown, inference optimizations, evaluation tasks, and identified limitations with suggested improvements.
KV CacheLLMMQA
0 likes · 9 min read


AIWalkerSep 24, 2025 · Artificial Intelligence
The article outlines four key innovation avenues—architecture redesign, task expansion, information fusion, and paradigm shift—highlighting recent works such as Mr. DETR, Grounding DINO 1.5, SM3Det, and RoboFusion, and offers a curated list of 176 cutting‑edge object‑detection papers with code and datasets for free.
Object Detectiondeep learningmodel architecture
0 likes · 8 min read


ArchitectSep 16, 2025 · Artificial Intelligence
This article introduces the Transformer architecture, explaining its attention mechanism, encoder‑decoder design, training and inference processes, and why it surpasses RNN‑based models, while also covering common applications and variations in natural language processing.
AttentionNLPTransformer
0 likes · 13 min read

This article provides a comprehensive technical comparison between Mixture of Experts (MoE) and the newly proposed Mixture of Recursion (MoR) architectures, covering design principles, parameter efficiency, inference latency, training stability, routing mechanisms, hardware deployment considerations, and suitable application scenarios.
Hardware DeploymentMixture of Recursioninference performance
0 likes · 13 min read

The paper introduces UniConvNet, a novel convolutional architecture that expands the effective receptive field (ERF) of ConvNets without breaking the asymptotically Gaussian distribution (AGD), achieving superior accuracy‑parameter and accuracy‑FLOPs trade‑offs across image classification, detection, and segmentation benchmarks.
Convolutional Neural NetworksEffective Receptive FieldUniConvNet
0 likes · 9 min read

LongCat‑Flash is a 560‑billion‑parameter Mixture‑of‑Experts LLM that combines a dynamic zero‑computation expert design, shortcut‑connected MoE communication, variance‑aligned scaling, and a three‑stage agent‑centric pre‑training pipeline, delivering over 100 TPS on H800 GPUs at a cost of $0.70 per million tokens.
Inference OptimizationLongCat-FlashTraining Infrastructure
0 likes · 23 min read

LongCat‑Flash‑Chat, Meituan’s newly open‑sourced 560B MoE model, outperforms leading LLMs on agent tool use and instruction following benchmarks, introduces zero‑computation experts and shortcut‑connected MoE for higher throughput, and demonstrates strong programming and reasoning abilities across diverse evaluation tasks.
Meituan AIZero Computation Expertsbenchmark performance
0 likes · 12 min read


Qborfy AIAug 8, 2025 · Artificial Intelligence
This article explains how the Transformer model replaces sequential RNN processing with parallel self‑attention, detailing its core components, positional encoding, encoder‑decoder workflow, industry impact, and surprising facts such as training speed gains and energy efficiency.
AISelf-AttentionTransformer
0 likes · 5 min read

This article analyzes the surprising design choices of the rumored GPT‑OSS 120B model, explaining the rationale behind a 64‑dimensional attention head, the equal hidden and intermediate sizes, and other quirky parameters such as MLP bias and KV‑sink SWA, backed by theoretical formulas and empirical benchmarks.
Attention HeadGPT-OSSMLP Ratio
0 likes · 13 min read

This article examines the evolution of large language model architectures over the past seven years, comparing key design choices such as Multi‑Head Latent Attention, Grouped‑Query Attention, Mixture‑of‑Experts, sliding‑window attention, normalization placement, and optimizer variants across models like DeepSeek V3, OLMo 2, Gemma 3, Llama 4, Qwen 3, SmolLM 3, and Kimi 2.
AI researchLLM comparisonattention mechanisms
0 likes · 30 min read

This comprehensive guide walks you through the evolution of AI model architectures—from CNNs and RNNs to Transformers and GANs—explaining their core concepts, typical use cases, and how to select, train, and fine‑tune pre‑trained models using practical code examples.
AIPretrained ModelsPython
0 likes · 25 min read

MiniMax’s latest M1 model, unveiled after a $300 million funding round, showcases a 4.56‑trillion‑parameter hybrid‑expert architecture with lightning attention, supporting up to one million tokens, and leverages reinforcement‑learning techniques to enhance long‑context handling, inference efficiency, and system‑2 reasoning capabilities.
AI scalingLong Contexthybrid attention
0 likes · 16 min read

This article provides a comprehensive technical analysis of Meta's Llama series, tracing the evolution from Llama‑1 through Llama‑2 to Llama‑3, detailing model architectures, training data pipelines, optimization methods, benchmark results, and the broader impact on the open‑source AI community.
AI researchLlamalarge language models
0 likes · 25 min read


Qborfy AIJul 1, 2025 · Artificial Intelligence
This article explains the fundamentals of convolutional neural networks (CNNs), detailing their definition, advantages over fully connected networks, architectural components such as input, hidden, and output layers, key operations like convolution, pooling, and activation, and showcases practical applications and notable insights.
CNNComputer Visionartificial intelligence
0 likes · 5 min read

This article examines DeepSeek’s various versions, detailing their core modules, underlying principles, architectural diagrams, and performance metrics, offering practical guidance for enthusiasts, professionals, and practitioners while inspiring further exploration of artificial intelligence innovations.
DeepSeekTech Overviewartificial intelligence
0 likes · 2 min read

The article provides a detailed technical overview of DeepSeek's flagship large language models, DeepSeek‑V3 and DeepSeek‑R1, describing their MoE architecture, training frameworks, reinforcement‑learning based fine‑tuning, inference optimizations, and the broader impact of these innovations on the AI landscape while also promoting related books and resources.
AIDeepSeeklarge language model
0 likes · 10 min read

This article provides a detailed Chinese‑to‑English summary of Andrej Karpathy’s 7‑hour LLM tutorial, covering chat process analysis, tokenization, pre‑training data pipelines, model architecture, training strategies, post‑training fine‑tuning, reinforcement learning, chain‑of‑thought reasoning, and current industry applications.
AILLMPretraining
0 likes · 25 min read

A new study introduces the lightweight “Canon” layer for large language models, showing how it improves information flow, inference depth, and scalability across Transformers, linear attention, and state‑space architectures, while offering a controlled synthetic pre‑training benchmark for deeper architectural analysis.
AI researchMambacanonical layer
0 likes · 11 min read

The article provides a comprehensive technical walkthrough of Qwen’s large‑model series, covering data preparation, tokenization, model tweaks, training settings, RLHF pipeline, Code‑Qwen specifics, Qwen2 and Qwen3 architectural changes, scaling‑law experiments, and detailed source‑code analysis with illustrative diagrams.
MoEQwenRLHF
0 likes · 7 min read

DeepSeek, a Chinese AI startup, offers open‑source large language models—DeepSeek‑V3 for general tasks and DeepSeek‑R1 for intensive reasoning—featuring MoE, MLA, low‑cost training, and competitive performance against OpenAI’s GPT‑4o, while providing detailed usage guidance and cost analysis.
AI inferenceDeepSeekcost efficiency
0 likes · 21 min read


ArchitectMar 10, 2025 · Artificial Intelligence
This article analyzes DeepSeek’s latest large‑model breakthroughs, covering the MLA attention compression, GRPO alignment algorithm, MoE load‑balancing redesign, multi‑stage training pipelines, reinforcement‑learning tricks, and performance comparisons with GPT‑4o‑Mini and Llama 3.1, highlighting both strengths and remaining challenges.
AI TrainingDeepSeekGRPO
0 likes · 19 min read

DeepSeek’s rapid rise is examined through its RL‑based Time Scaling paradigm, cost‑effective architecture, innovative training pipeline, open‑source strategy, and security challenges, highlighting how these breakthroughs disrupt traditional AI model development, lower resource demands, and influence industry dynamics.
AI model trainingDeepSeekcost-efficient AI
0 likes · 13 min read

This article examines DeepSeek’s various versions, detailing their core modules, underlying principles, architecture diagrams, and performance metrics, while illustrating the internal logic and advantages of each model to guide enthusiasts, professionals, and practitioners toward deeper AI innovation insights.
AIDeepSeekTechnology
0 likes · 4 min read

The article analyzes DeepSeek’s recent breakthroughs—including its Mixture‑of‑Experts architecture, cost‑effective inference optimizations, high‑accuracy multimodal processing, and open‑source collaboration—while also offering a curated bundle of technical e‑books covering AI chips, networking, storage, and more.
DeepSeekInference Optimizationartificial intelligence
0 likes · 4 min read

This article dissects the Multi‑Token Prediction (MTP) technique used in DeepSeek‑R1, contrasting it with traditional next‑token prediction, detailing Meta’s MTP design, DeepSeek’s adapted architecture, loss weighting, and why MTP is applied only during training to boost efficiency and model capability.
DeepSeekMTPMulti-Token Prediction
0 likes · 9 min read


JavaEdgeFeb 8, 2025 · Artificial Intelligence
This article provides a detailed technical analysis of DeepSeek's R1 large language model, covering its background, architecture, training methods, hardware optimizations, performance claims, user impressions, deployment options, and the challenges of reproducing its results.
AI TrainingDeepSeekGPU Cost
0 likes · 16 min read

The article provides a detailed analysis of DeepSeek’s model performance across language, reasoning, and code generation benchmarks, its cost‑effective training methods, novel architecture innovations, the team’s expertise, and the broader impact these factors may have on accelerating AI innovation and reshaping industry competition.
AI benchmarksAI industry impactDeepSeek
0 likes · 18 min read

The article analyzes the US‑China large‑model race, detailing algorithmic and architectural strengths of OpenAI, Google and Microsoft versus Chinese innovations like Doubao 1.5, MiniMax‑01 and Vidu, and projects a timeline from 2025 to 2033 for China to close the gap.
AI competitionChinaUnited States
0 likes · 12 min read


NewBeeNLPJan 17, 2025 · Artificial Intelligence
This comprehensive survey examines the foundations, tokenization techniques, model architectures, training paradigms, evaluation benchmarks, and open challenges of multimodal next‑token prediction (MMNTP), offering researchers a clear roadmap for future advances in multimodal AI.
Next Token PredictionTraining Paradigmsevaluation
0 likes · 9 min read

This article presents Baidu's latest advances in multimodal large models, detailing their capabilities, architectural evolution, real‑world applications, and the open‑source PaddleMIX toolkit that streamlines data processing, training, fine‑tuning, and high‑performance inference for developers.
AI applicationsPaddleMIXdata processing
0 likes · 20 min read

This article details the end‑to‑end process of pre‑training, fine‑tuning, and evaluating a 1‑billion‑parameter Chinese LLM named Steel‑LLM on limited hardware, covering data collection, pipeline design, training framework choices, architectural tweaks, performance results, and practical lessons for resource‑constrained developers.
LLMPretrainingdata pipeline
0 likes · 18 min read

This article reviews the necessity, benefits, and practical implementations of multi-task learning in e‑commerce search, detailing model selection, architecture extensions such as ESMM and ESM², and future directions for handling user behavior sequences and multi‑objective optimization.
ESMMconversion rate predictiondeep learning
0 likes · 13 min read

This article reviews recent advances in multimodal large language models, covering their background, architectural components, training strategies, application scenarios, evaluation benchmarks, team research on hallucination mitigation and long‑video understanding, and outlines promising future research directions.
evaluation benchmarksfuture researchmodel architecture
0 likes · 15 min read

This article presents 360's comprehensive exploration of image‑text multimodal large models, covering background concepts, research routes, three generations of model development, proprietary architectures like SEEChat, 360VL and Inner‑Adaptor, and real‑world AI applications across various products and services.
AI applicationsmodel architecturevision-language
0 likes · 19 min read


NewBeeNLPOct 21, 2024 · Artificial Intelligence
This article analyzes the polarization issues in industrial Mixture‑of‑Experts (MoE) frameworks, explains expert collapse, degradation, and under‑fitting, and details the HOME model’s input types, architectural innovations, normalization, gating mechanisms, and related DICE‑BN insights.
Expert NormalizationGating Mechanismsmixture of experts
0 likes · 10 min read


ArchitectSep 26, 2024 · Artificial Intelligence
This article provides a detailed technical analysis of OpenAI’s o1 model, exploring its enhanced logical reasoning, the likely use of reinforcement learning with hidden chain‑of‑thought generation, multi‑model architecture, training data pipelines, reward modeling, and how these innovations could reshape AI safety and scaling strategies.
AI safetyLLMOpenAI o1
0 likes · 43 min read

This article presents NetEase Cloud Music's multi‑scenario recommendation modeling work, detailing background, overall system architecture, key modules, modeling goals, technical difficulties, performance improvements, future outlook, and a comprehensive Q&A session that addresses practical deployment challenges.
AB TestingAINetEase Cloud Music
0 likes · 14 min read

The article provides an in‑depth analysis of LLaMA 3 405B, covering its dense Transformer architecture, three‑stage pre‑training (initial, long‑context, annealing), iterative post‑training with RM‑guided rejection sampling, the decision against MOE, and the broader implications for both large and small model development.
405BPretrainingmodel architecture
0 likes · 17 min read

This article dissects Meta’s Llama 3 405‑billion‑parameter model, covering its dense Transformer design, data‑mixing strategy, two‑stage scaling‑law prediction, 4‑D parallelism, custom hardware clusters, training schedules, post‑training alignment methods, and the extensive evaluation results that benchmark it against leading LLMs.
AI infrastructureLlama-3distributed training
0 likes · 56 min read

The article provides a detailed technical overview of Gemma 2, covering its decoder‑only transformer design, novel attention mechanisms, logit soft‑capping, RMSNorm, knowledge‑distillation training on trillions of tokens, extensive pre‑training infrastructure, and benchmark evaluations that demonstrate its competitiveness against larger proprietary models.
AIGemma 2benchmark evaluation
0 likes · 14 min read


NewBeeNLPJun 7, 2024 · Artificial Intelligence
In a recent round‑table, experts debated the validity of scaling laws, the role of synthetic and semi‑synthetic data in overcoming data scarcity, explored alternatives to Transformers such as RNN‑based models and MOE, and examined techniques for handling long‑context inference efficiently.
Scaling Lawsmixture of expertsmodel architecture
0 likes · 12 min read

Meta's Llama series has progressed from the 7‑65B Llama‑1 in early 2023 to the 8B and 70B Llama‑3 in 2024, scaling token counts from 1 T to over 15 T, adopting decoder‑only Transformers with RMSNorm, SwiGLU, RoPE and GQA, and adding supervised fine‑tuning, RLHF and DPO, resulting in state‑of‑the‑art benchmark performance and a vibrant open‑source ecosystem.
AILlamalarge language models
0 likes · 25 min read


NewBeeNLPMar 27, 2024 · Artificial Intelligence
This article provides a comprehensive technical overview of Meta's Llama 2 series, covering its architectural upgrades such as Group Query Attention, the pre‑training dataset and hyper‑parameters, loss behavior, benchmark comparisons, and the supervised fine‑tuning pipeline with safety considerations.
AILlama 2Pretraining
0 likes · 11 min read


21CTOMar 18, 2024 · Artificial Intelligence
Elon Musk’s AI startup xAI has open‑sourced its 314‑billion‑parameter Grok‑1 model, detailing its Rust‑based, JAX‑powered architecture, extensive parameter count, training data limits, licensing terms, hardware requirements, and community reactions, offering developers unprecedented access to a competitive large‑language‑model framework.
AIGrok-1JAX
0 likes · 9 min read

Bilibili’s self‑supervised video super‑resolution system upgrades low‑resolution streams to 4K by using three parallel degradation‑branch networks—texture‑enhancing, line‑recovering, and noise‑removing—tailored to anime, game, and real‑world content, delivering sharper edges, finer textures, and measurable quality gains across its online playback pipeline.
AIBilibilideep learning
0 likes · 16 min read

This article presents a comprehensive overview of financial large language models, covering their unique characteristics, construction methods, layered technical architecture, evaluation strategies, and real‑world use cases such as quality inspection, AIGC‑driven material generation, sales‑lead mining, and knowledge‑graph‑enhanced intelligent Q&A.
Data Engineeringfinancial AImodel architecture
0 likes · 14 min read

This article walks through the core LLaMA implementation in HuggingFace’s Transformers library, detailing the inheritance hierarchy, configuration defaults, model initialization, embedding and stacked decoder layers, the RMSNorm‑based attention and MLP modules, and the forward pass that produces normalized hidden states.
PyTorchartificial intelligencedeep learning
0 likes · 14 min read

This article reviews the LLaMA large‑language‑model series, covering its background, architectural innovations such as Add&Norm, SwiGLU, and RoPE, a known reversal‑curse bug, and provides step‑by‑step MindSpore Transformers code for model configuration, inference, and pipeline usage while previewing the upcoming LLaMA‑2 session.
AILlamaMindSpore
0 likes · 6 min read

The article critically examines GPT‑4 and similar transformer models, highlighting persistent hallucinations, outdated knowledge, insufficient domain coverage, lack of planning and memory, and proposes architectural extensions inspired by fast‑slow thinking and differentiable modules to overcome these fundamental constraints.
AI limitationsGPT-4Hallucination
0 likes · 24 min read

This article presents Alibaba Health's exploration of explainable recommendation algorithms, covering business context, data preparation, feature extraction and encoding, model architecture combining selection and prediction components, experimental offline and online results, and a detailed Q&A on implementation challenges and future directions.
AIAlibaba Healthexplainable recommendation
0 likes · 12 min read

This article presents a comprehensive overview of multimodal pre‑training, describing its motivation, architecture choices, large‑scale Chinese image‑text dataset construction, training optimizations, performance benchmarks, downstream applications, and a Q&A session that highlights practical deployment considerations.
Computer VisionPretrainingdeep learning
0 likes · 16 min read

This article describes how a second‑hand e‑commerce company built a fast, modular image‑classification pipeline using small binary classifiers, efficientNet‑b0, and active‑learning‑driven data annotation to detect prohibited items while keeping inference latency under 200 ms and reducing labeling costs dramatically.
AIActive Learningimage classification
0 likes · 10 min read

This article presents Douyu's intelligent risk‑control system for live streaming, detailing the operational, activity, traffic, account, transaction and content safety challenges, the multi‑layer algorithm architecture, and the evolution of models for spam detection, risk scoring, gang identification, behavior sequencing, device fingerprinting, and interpretability.
Risk Controlartificial intelligencefraud detection
0 likes · 13 min read

This article introduces the fundamentals of transfer learning and federated transfer learning, explains domain adaptation for sentiment analysis, presents two illustrative examples—mid-level image feature transfer and text-to-image transfer—and outlines the model architecture and loss functions of federated transfer learning frameworks.
domain adaptationmodel architectureprivacy
0 likes · 14 min read

The ICCV 2019 Lightweight Face Recognition Challenge attracted 292 teams and defined four strict FLOP‑ and size‑limited protocols for image and video recognition, with champions employing near‑30 GFLOP EfficientNet‑style backbones, novel loss functions, frame‑fusion, and knowledge‑distilled VarGNet models to balance accuracy and computational budget.
Computer VisionICCV ChallengeLightweight Face Recognition
0 likes · 8 min read

This article explains the design and evaluation of RE2, a lightweight yet expressive text‑matching framework for FAQ‑style chatbots, detailing its five‑layer architecture, block‑wise residual connections, experimental results on SNLI, MultiNLI, SciTail, Quora and WikiQA datasets, and its significant performance improvements in Alibaba’s DingXiaoMi service.
FAQ chatbotNLPdeep learning
0 likes · 13 min read

This article reviews the concept of residual networks, explains their working principle and data‑flow interpretation, discusses why they improve deep models, analyzes path‑length effects on gradients, and surveys various residual block designs and practical takeaways.
ResNetensemblegradient
0 likes · 9 min read

This article traces the evolution of convolutional neural networks from the pioneering AlexNet through VGG, Inception, ResNet, Inception‑v4, Inception‑ResNet and ResNeXt, highlighting architectural innovations, performance gains, and the underlying biological inspirations that shaped modern deep learning models.
AlexNetCNNComputer Vision
0 likes · 13 min read
