Ximalaya AI Cloud leverages DeepRec’s Embedding Variable to elastically manage high‑dimensional sparse features with low collision, supporting admission/eviction, multi‑level storage and minute‑level incremental model updates, which together boost GPU utilization, halve training time and improve recommendation CTR by 2‑3 % while maintaining latency.
The article explains why large language models generate hallucinations—due to data errors, training conflicts, and inference uncertainty—and outlines data‑cleaning, model‑level feedback, knowledge augmentation, constraint techniques, and post‑processing methods such as the “Truth‑seeking” algorithm to mitigate the issue.
This article introduces the open‑source Pai‑Megatron‑Patch tool from Alibaba Cloud, explains its non‑intrusive patch architecture, enumerates supported models and features such as weight conversion, Flash‑Attention 2.0, FP8 training with Transformer Engine, and provides detailed command‑line examples for model conversion, pre‑training, supervised fine‑tuning, inference, and RLHF reinforcement learning pipelines.
This guide compares Stable Diffusion fine‑tuning methods, shows why LoRA offers the best size‑and‑speed trade‑off, and provides a step‑by‑step workflow—from dataset collection and preprocessing to parameter tuning, 20‑minute training on a single GPU, and practical tips for successful custom model generation.
This article introduces the YOLOv8 object‑detection framework on the Ultralytics platform, covering environment setup, command‑line and Python APIs for inference, model‑file options, result interpretation, data annotation, training procedures, and exporting models to various deployment formats.
This step‑by‑step tutorial shows how to set up a GPU environment, prepare custom text data, train a tokenizer, configure and train a GPT‑2‑based large language model, test its generation, and run the entire pipeline using only 61 lines of Python code.
This article explains what GPT is, how it is trained, its strengths and limitations, the various risks it poses, and provides practical guidance on safely adopting large language models like ChatGPT within development teams, including code‑level analysis examples.
The article examines how AI-generated misinformation spreads across platforms—from misleading answers on Bing and Stack Overflow to fabricated news stories—highlighting the resulting contamination of online information, the risks to model training, and emerging efforts to detect and curb such low‑quality AI output.
This comprehensive guide walks you through the hardware requirements, environment deployment, key parameters, prompt techniques, ControlNet integration, model download and installation, as well as style and character training for Stable Diffusion, providing practical code snippets and visual examples for each step.
The article explains that reproducing ChatGPT in China is not just a matter of funding but requires extensive expertise in large‑scale language model training, massive compute resources, optimized cloud infrastructure, and deep AI research, as demonstrated by Alibaba's DAMO Academy efforts.
This article provides an overview of ChatGPT’s rapid rise, OpenAI’s founding, the evolution of GPT models up to GPT‑3, the data‑driven training process, model capabilities and limitations, and practical guidance for users, highlighting the interplay between open‑source research and commercial deployment.
This article summarizes how Ant Group introduced Alluxio to address storage I/O, capacity, and latency challenges in large‑scale model training, detailing stability improvements through worker‑register follower and master migration, performance gains via follower‑only reads, and horizontal scaling using metadata sharding and multi‑cluster deployment.
This article explains how ChatGPT builds on GPT‑3, describes the supervised‑plus‑reinforcement learning (RLHF) pipeline that fine‑tunes the model, compares model capability with consistency, and discusses the performance evaluation and remaining limitations of large language models.
This article outlines Baidu Intelligent Cloud's comprehensive, low‑cost solution for autonomous‑driving data pipelines—from road data collection and compliance, through annotation, management, and model training, to simulation—highlighting the platform's tools, services, and security measures that accelerate development.
This article examines the use of Mixture‑of‑Experts (MoE) sparse training for GPT models, detailing the architecture, training and inference efficiency gains, experimental comparisons with dense models, custom routing algorithms, and step‑by‑step deployment on Alibaba Cloud AI platforms.
The talk outlines Baidu’s Baige AI IaaS solution for autonomous driving, detailing a low‑cost, high‑efficiency cloud stack that accelerates data access fivefold, boosts model training speed up to 391 %, cuts inference latency by 90 %, reduces simulation costs by 60 %, and explains the underlying storage, compute, container and GPU virtualization technologies.
This article explains how a unified AI platform enables rapid, data‑driven creation, training, deployment, and evaluation of OCR models for visually distinct text regions such as seals, meter readings, license plates, and VIN codes, while minimizing hardware and annotation costs.
This article presents a hybrid pipeline combining traditional computer vision techniques and deep learning models to automatically extract and verify text background colors in UI automation screenshots, effectively addressing challenges like limited training data and complex borders to significantly reduce manual inspection costs while achieving high accuracy and robustness in production environments.
Yanxuan built a Kubeflow‑based machine‑learning platform that unifies data preprocessing, feature engineering, model training, validation, and deployment, using Smart‑jobs, Smart‑Infer, Smart‑backend, Airflow pipelines, Jupyter notebooks, and Istio‑enhanced inference services to boost algorithm engineers’ efficiency and integrate with Kubernetes, HDFS, and Hive.
Airflow orchestrationAlgorithm DevelopmentInference Service
This article provides a thorough collection of commonly used PyTorch code snippets covering environment setup, reproducibility, GPU configuration, tensor manipulation, model building, data preprocessing, training and evaluation loops, custom loss functions, regularization techniques, learning‑rate scheduling, checkpointing, and practical tips for efficient deep‑learning development.
The article details Alibaba’s end‑to‑end online‑learning pipeline for real‑time ranking in the Taobao home‑decor channel, covering UT log parsing, full‑feature extraction, ODL sample creation, xDeepCTR model training, and deployment, which yielded up to 7.8% CTR improvement and demonstrates the value of rapid model adaptation.
This article presents a comprehensive study on applying AI and deep‑learning techniques to mobile UI testing, covering background challenges, feasibility research, abnormal sample construction, model design, training, evaluation, and future directions for intelligent test automation.
The article introduces xeasy-ml, a Python machine‑learning framework, explains its installation requirements, demonstrates how to initialize a project, run training, view result artifacts, and provides detailed instructions for online prediction using configuration files and code examples.
The article outlines how MNN Workbench, Alibaba’s open‑source edge‑AI platform, integrates professional training capabilities, cloud‑based PAI‑DLC resources, multi‑window debugging, and visual Git Flow to streamline end‑to‑end model development, deployment, and iteration for developers of varying expertise.
This article presents the challenges of big‑data architecture in finance and introduces three integrated solutions—MMR cloud‑based architecture, the Honghu data‑lake management and analysis platform, and the Yichuang model‑training monitoring system—detailing their design, governance, and future outlook.
Cloud ArchitectureFinancial TechnologyModel Training
This article presents Du Xiaoman Financial's big‑data architecture challenges and three integrated solutions—MMR cloud‑based data framework, the Honghu data‑lake management platform, and the Yichuang model‑training monitoring system—detailing their design, governance, low‑threshold usage, and future outlook.
This article presents a deep‑learning approach that transforms recorded audio into spectrograms, trains a noise‑prediction network (e.g., ResNet, U‑Net, LSTM) to estimate environmental noise, subtracts it in the frequency domain, and reconstructs a cleaner signal for more accurate audio‑device quality assessment.
TDQA is a no‑reference, deep‑learning video quality assessment algorithm designed for live‑streaming, built on a large subjectively annotated dataset and an end‑to‑end architecture with fine‑tuned backbones, achieving state‑of‑the‑art accuracy and sub‑second inference for real‑time quality monitoring and pipeline optimization.
Learn how to use Python to send real-time progress updates and completion alerts via email for long-running tasks such as model training, data processing, and financial modeling, by leveraging the email, smtplib, and MIME libraries to build customizable notifications with images and attachments.
This article introduces Baidu EasyDL's zero‑threshold AI training camp, focusing on an automatic‑driving scene image‑segmentation case study, covering platform features, model selection, data collection, cleaning, annotation, smart labeling, augmentation, deployment, and a live Q&A session.
This article provides a comprehensive engineering overview of federated learning, covering its core distributed‑learning concept, Docker‑based deployment, detailed training‑service architecture with validation, scheduling, metadata, and model‑management components, as well as a complete inference framework and workflow for production use.
The article details the end‑to‑end architecture of Toutiao's massive recommendation platform, covering system overview, content and user feature extraction, model training, recall strategies, evaluation methodology, and content safety mechanisms, while highlighting practical challenges and engineering solutions.
Content AnalysisModel TrainingRecommendation System
This article provides a step‑by‑step tutorial on preparing data, pre‑training the SPTM language model, fine‑tuning a text‑classification model, generating predictions, and creating a submission file for the 58.com AI algorithm competition using the open‑source qa_match toolkit.
This article proposes a meta‑knowledge transfer framework for AutoML systems, detailing a four‑layer architecture, methods for collecting and updating structured model meta‑knowledge, and strategies that use this knowledge to guide hyper‑parameter search and early‑stop training, thereby improving efficiency and reducing resource consumption.
Sunfish is Youzan’s integrated AI platform that unifies visual drag‑and‑drop model training, notebook‑based algorithm development, automated model management and one‑click publishing with a low‑latency, high‑availability “small‑box” inference service, enabling end‑to‑end deep‑learning workflows from data exploration to online recommendation and risk‑control deployment.
This tutorial walks through creating a TI-ONE Notebook instance, importing competition data, training a TensorFlow model, and uploading results, providing a complete end‑to‑end workflow for participants of the 2020 Tencent Advertising Algorithm Competition.
This tutorial walks you through creating a TI‑ONE project, ingesting competition data, configuring and training a decision‑tree model with built‑in operators, running the workflow, and downloading and uploading the result files for the 2020 Tencent Advertising Algorithm Competition.
This article presents the overall design, multi‑dimensional control mechanisms, auxiliary modules, and evolutionary processes of the quality control system used in WeChat's Look feature, detailing data lifecycle, model training, generalization, transfer learning, and continuous anti‑abuse strategies.
The article describes how Alibaba's XiaoMi AI platform constructs a closed‑loop pipeline—from data collection and annotation to model training, evaluation, and real‑time deployment—using multi‑dimensional data processing, visualization, and Spark‑based engines to accelerate iterative improvements and address operational pain points.
This article describes a complete pipeline for recommending automated test scripts for APK releases, covering CSV data preprocessing, feature encoding, tokenization with pkuseg and jieba, and training various machine‑learning models such as LDA, word2vec, XGBoost, deep neural networks, and multi‑label classifiers to predict script execution order.
This article describes a standardized workflow for feature collection, configuration, processing, and model training/prediction in large‑scale recommendation systems, using CSV‑based definitions and code generation to ensure consistency between offline training and online serving while reducing manual coding effort.
This article describes the challenges of large‑scale model training for Snowball’s feed recommendation, and details a series of engineering optimizations—including GPU acceleration, multi‑threaded data preparation, TFRecord conversion, compression, and batch‑map reordering—that increased training throughput from 6 k to over 20 k samples per second while reducing CPU and I/O bottlenecks.
At ACL 2019, Didi unveiled DELTA, an open‑source TensorFlow‑based training framework that unifies NLP and speech tasks, offers configurable pipelines, benchmark models, and seamless deployment, enabling AI developers to quickly move from research to production while leveraging Didi’s extensive open‑source ecosystem.
The article explains how large‑scale machine learning and AutoML are applied to search advertising click‑through‑rate (CTR) prediction, covering problem definition, feature generation, model training, optimization methods, distributed systems, and recent advances in AutoML with practical case studies.
The Nanjing University team '每天队员都想改一次名字' shares their winning approach in Tencent Advertising Algorithm Competition, covering data processing, feature engineering, model training techniques, and alternative optimization targets for AUC improvement, and discuss lessons learned from previous year's champion experience.
The article outlines the seven‑stage machine‑learning pipeline—from data acquisition and preparation through model selection, training, evaluation, parameter tuning, and deployment—highlighting the critical role of feature engineering, which typically consumes over 80% of project time.
The article explains that while machine learning has advanced with abundant courses, textbooks, and frameworks, engineers still face hard debugging problems due to algorithmic, implementation, data, and model dimensions, leading to exponential difficulty and long feedback loops that demand intuition and systematic testing.
This article presents a comprehensive, practical guide to applying machine learning in industry, covering problem modeling, data preparation, feature extraction, model training, and optimization, illustrated with a DEAL transaction amount forecasting case study.
This article outlines a complete industrial machine‑learning workflow—starting with problem modeling, through data preparation, feature extraction, model training, and ending with model optimization—illustrated with a real‑world DEAL revenue‑prediction case and practical tips for handling data, features, and model selection.
This article outlines how Meituan applies machine learning to industrial challenges by detailing the full workflow—from problem modeling and data preparation to feature engineering, model training with algorithms like Logistic Regression and GBDT, and optimization techniques that address underfitting and overfitting for large‑scale deployment.
This guide walks through Meituan’s end‑to‑end offline ML workflow—from problem modeling and data preparation, through feature engineering and normalization, to model selection, training optimization, evaluation, and iterative improvement—emphasizing business alignment, data quality, and practical diagnostics for real‑world deployment.
Model Trainingfeature engineeringindustrial application