Self‑Variable, the leading Chinese embodied‑intelligence startup, completed four rounds of financing worth over 200 billion RMB, unveiled its world‑unified‑model WALL‑B and open‑source models, and began deploying home robots, marking a pivotal shift from early‑stage R&D to commercial rollout in the Greater Bay Area.
TimechoAI, the cloud service built on the Timer time‑series large model, offers the latest SOTA model (Timer‑3.5) alongside classic baselines, supports multiple data input methods, covariate integration, and API/SDK access, and invites industrial and IoT teams to test its predictive maintenance, production optimization, energy load forecasting, and anomaly detection capabilities through a simple invitation process.
Xiaomi unveils OneVL, an open‑source stepwise latent language‑vision reasoning framework that unifies VLA, world‑model and latent inference, delivering higher accuracy than explicit CoT and inference speed comparable to answer‑only models, with SOTA benchmark results across multiple autonomous‑driving tests.
The article introduces time‑series data, its ubiquitous examples, the challenges of traditional small models, and proposes a universal time‑series large model that simplifies data preparation and model building, ultimately enabling more efficient and stable industrial AI solutions, now offered as a cloud service.
This article systematically presents ten large‑model training and fine‑tuning methods—from full‑parameter finetuning to parameter‑efficient PEFT—detailing their principles, suitable scenarios, step‑by‑step workflows, code examples, and practical selection guidance for AI product managers.
The article walks through model training, validation, ensemble learning, and deployment from an AI product manager’s viewpoint, using a churn‑prediction case to illustrate decision boundaries, metric choices, industry‑specific algorithm trade‑offs, cost considerations, and practical serving options.
The article details how Knora 4.0 integrates ontology with large‑model AI to create a reusable, extensible enterprise AI platform that mitigates hallucination, stabilises output, and enables autonomous end‑to‑end execution, illustrated with LED production line case studies, architectural breakdowns, and a roadmap for future intelligent agents.
An AI Agent is a large‑model system that can perceive its environment, plan steps, invoke tools, and remember past interactions to autonomously achieve user‑specified goals, distinguishing it from simple chatbots that only answer questions.
This guide walks you through interpreting training logs and loss curves, diagnosing common issues such as oscillation, under‑fitting, and over‑fitting, and applying concrete adjustments to learning rate, LoRA settings, batch size, and epochs, with scenario‑specific strategies to turn a novice into a tuning expert.
LongCat-Next is a 68.5‑billion‑parameter discrete‑native autoregressive multimodal model that tokenizes images, audio and text, challenges the belief that visual tokenization loses detail, matches specialized models on fine‑grained tasks, and demonstrates that joint understanding‑generation training can even improve generation quality.
Xiaomi introduces MiMo-V2-TTS, a self‑developed speech‑synthesis large model that combines a custom audio tokenizer, multi‑codebook architecture, massive pre‑training on over a hundred million hours of data and multi‑dimensional reinforcement learning to deliver fine‑grained style control, dialect support, role‑play and high‑quality singing, aiming to give AI agents expressive, human‑like voices.
The STAPO (Spurious‑Token‑Aware Policy Optimization) algorithm, introduced by Tsinghua University's iDLab and Didi's Deep Sea Lab, tackles policy‑entropy instability and performance oscillation in reinforcement‑learning fine‑tuning of large models by mathematically analyzing token collision probability, defining spurious tokens, and applying a Silencing Spurious Tokens mechanism that yields state‑of‑the‑art results on multiple math‑reasoning benchmarks.
This article presents a comprehensive methodology for selecting AI agent scenarios, explains the economic benefits of agent deployment, details the technical architecture—including domain large models, knowledge bases, planning strategies, and RL‑based scheduling—and illustrates how these components are applied to insurance product design, pricing, fulfillment, and risk control to drive scale and profit.
Microsoft’s AdNanny replaces a forest of specialized offline models with a single 671 B LLM, using a three‑stage data factory to generate reasoning‑rich corpora, dynamic task re‑weighting, RL‑based metric alignment, and a hybrid 31‑pipeline‑parallel architecture that halves compute cost while boosting performance on core ad‑ranking tasks.
This guide explains the background of speech‑to‑text technology, introduces the large‑model streaming speech recognition API, walks through obtaining an API key, and provides detailed Go and Python code for establishing a WebSocket connection, sending full‑client and audio‑only requests, and parsing server responses.
The article examines why most AI agents operate as workflow‑driven systems due to the reliability, accuracy, execution and cost limits of large models, and offers practical guidance on designing, evaluating, and iterating effective workflow agents while acknowledging their constraints and future prospects.
This guide walks through Coze's large‑model configuration—including model selection, generation diversity, input/output parameters, and persona templates—demonstrates a complete e‑commerce chatbot setup, and details two plugin integration methods with three concise best‑practice rules for effective agent development.
The article reviews the evolution of e‑commerce search, explains why traditional multi‑stage pipelines struggle with relevance and personalization, and details OneSearch’s end‑to‑end large‑model design—hierarchical quantization encoding, integrated user profiling, and preference‑aware ranking—that achieved up to 27% model‑utilization and double‑digit CTR, CVR, and order growth in production.
This article outlines a systematic engineering workflow for creating professional domain fine‑tuning datasets for large models, covering data processing, validation, optimal sample size, industrial‑environment practices, and special considerations for reinforcement‑learning based fine‑tuning.
The article explains how to estimate GPU memory and overall hardware requirements for fine-tuning and training large dense and MoE models, detailing calculations for full-parameter and LoRA approaches, scaling rules, and hidden costs relevant to interview assessments.
JD’s internal technical salon reveals the rapid evolution of large‑model AI tools, outlining design principles, current capabilities like JoyAgent, OxyGent and JoyCode, real‑world applications across office, code review, logistics, and future policy and multi‑agent visions, highlighting their impact on modern workplace productivity.
JD’s internal technical salon reveals the rapid evolution of large‑model AI tools, outlining design principles, current capabilities like JoyAgent, OxyGent and JoyCode, real‑world applications across office, code review, logistics, and policy outlook, while highlighting future multi‑agent and embodied intelligence trends.
The article summarizes Professor Long Mingsheng’s presentation on the Timer series of time‑series large models, detailing the three core challenges of industrial time‑series analysis, the evolution from statistical methods to generative models, and the technical breakthroughs of Timer 1.0, 2.0 and 3.0 that enable multi‑task, long‑context, and trillion‑scale forecasting for industrial digital transformation.
At the 2025 AI Innovation & Entrepreneurship Conference in Hangzhou, JD.com’s chief scientists unveiled a billion‑parameter time‑series large model and end‑to‑end inventory algorithms that dramatically boost demand forecasting, dynamic allocation, and overall supply‑chain efficiency, illustrating how AI can transform global logistics networks.
This article presents a curated list of cutting‑edge resources covering multimodal GraphRAG, knowledge‑graph‑driven large‑model applications in finance, healthcare, automotive, and more, offering insights into the evolving synergy between AI and knowledge graphs.
This report examines large model platforms as the engineering foundation for turning base models into business applications, outlines the full lifecycle toolchain, discusses key challenges such as technology selection and secure management, and provides practical case studies and strategic guidance for scalable, value‑driven deployment across industries.
At GAITC2025, JD Retail’s AI Infra lead Zhang Ke detailed the challenges of e‑commerce AI inference and introduced the xLLM edge‑cloud unified large‑model architecture, highlighting adaptive scheduling, offline unified scheduling, multi‑layer pipelines, and agent collaboration that boost performance, cut costs, and pave the way for future AI advancements.
The paper "Omniforce" from JD Exploration Research Institute presents a cloud‑edge collaborative AutoML system that uses model distillation, data governance, Bayesian training optimization, and cloud‑edge cooperation to reduce large‑model training costs by 70% and improve inference efficiency by an average of 30%, offering a reusable technical paradigm for scalable AI deployment.
Meng, a former Tencent and Alibaba engineer, led Dewu’s one‑stop large‑model training and inference platform, cutting integration costs, creating a shared GPU pool and CI/CD pipeline, building a Milvus vector‑database, and driving self‑directed learning that boosted business value, user experience, and set a roadmap for future RAG and cloud‑native optimizations.
On April 2, Baidu released its Wenxin X1 large model on the Qianfan platform, offering enterprise users and developers a multimodal, deep‑thinking AI with superior math, coding, and reasoning scores, low token‑price API access, batch inference, one‑click distillation, and rapid RAG/Agent application building.
Deploying large‑model applications can lead to unpredictable token consumption far beyond traditional web services, driven by factors such as model type, input/output length, caching, chain‑of‑thought, prompt design, and ecosystem dependencies, requiring comprehensive monitoring, preventive controls, and post‑incident optimizations.
This article analyzes how DeepSeek’s multimodal large model overcomes traditional digital‑twin simulation bottlenecks through dynamic modeling, generative data augmentation, and low‑cost open‑source architecture, enabling smarter city traffic, industrial design, and water‑resource management while reshaping the industry’s AI‑driven simulation landscape.
This manual introduces the fundamental concepts of the DeepSeek‑R1 model, explains large‑model and AIGC technologies, and provides practical guidance for selecting and efficiently using AI tools, helping readers grasp the deeper value of DeepSeek and related applications.
Huolala’s Wukong platform tackles the common “technology hype, implementation difficulty” dilemma of generative AI by unifying multimodal enterprise knowledge, enabling dynamic multi‑agent workflows, and providing low‑code tools, observability, and stable deployment across dozens of business scenarios.
This article examines the importance of domestic large‑model inference engines, compares Chinese and international chips, evaluates four architectural approaches, discusses practical challenges such as performance loss and model support, and outlines future expectations for high‑performance, heterogeneous‑chip inference solutions.
As AI accelerates, large‑model compute infrastructure becomes a cornerstone of digital transformation, with specialized accelerators, heterogeneous architectures, massive distributed clusters, intelligent scheduling, soaring costs, energy concerns, software‑hardware co‑design challenges, and data‑privacy issues shaping its future development.
The article details Ant Group’s AI‑driven security parallel plane and intelligent threat detection system, its DKCF‑based architecture, key modules for data correlation, unknown threat discovery, alarm reduction, and knowledge‑graph integration, and its recognition in the 2024 AI Pioneer Case Collection.
The talk outlines the evolution of intelligent customer service over three decades, explains how generative large-model AI like ChatGPT has raised service expectations, and presents Alibaba Cloud’s four-stage implementation—experience, efficiency, capability, and insight—through three concrete cases and a roadmap for SMEs to build their own smart service systems.
The article explains what AI large‑model integrated machines are, why they are needed for data‑sensitive sectors, and presents market data showing a rapidly growing demand in China that could reach a trillion‑yuan scale, especially as domestic alternatives replace imported GPUs.
The article traces the evolution of large‑model training and inference infrastructure from the early “black‑iron” era (2019‑2021) through the “golden” boom (2022‑2023) to the emerging “silver” phase (2024‑), highlighting key research breakthroughs, open‑source frameworks, hardware trends, market dynamics, and practical challenges for engineers entering the field.
The jointly authored 2024 Large Model Security Practice whitepaper, released at the World AI Conference, outlines a comprehensive safety framework covering security, reliability, and controllability, presents industry case studies, and proposes a five‑dimensional governance model to guide high‑quality development of large AI models.
This article presents a comprehensive knowledge map for developing large‑model applications, covering a four‑layer technical architecture, the full development lifecycle, core elements such as prompt engineering and model fine‑tuning, evaluation methods, and practical case studies, offering guidance for both enterprises and startups.
The article details Ant Group’s security parallel aspect fusion AI solution, selected as an exemplary case at the 2024 World Intelligent Industry Expo, explaining its multi‑dimensional data collection, large‑model integration, baseline construction, knowledge‑graph generation, and superior threat‑detection performance.
This article outlines the key characteristics of large‑model enterprise applications, compares them with consumer use cases, and presents a comprehensive engineering roadmap—including model selection, knowledge‑base integration, tool implementation, intent recognition, output control, high‑availability deployment, and ongoing optimization—to help practitioners effectively harness AI models in real‑world business environments.
At the iTechClub summit, Baidu unveiled Comate Code Knowledge Enhancement 2.0—the nation’s first AI‑driven code assistant with real‑time web retrieval, over 10,000 built‑in GitHub repositories, API‑call generation, test case creation, and deep private‑doc understanding, now producing 27% of Baidu’s daily new code and topping VSCode and JetBrains plugin ratings.
The article explains the scaling challenges of ever‑larger LLMs, introduces the MFU performance metric, surveys industry parallelism and memory‑saving techniques, and details Baidu’s AIAK‑LLM suite—including resource, component and acceleration layers—as well as concrete training and inference optimizations that raise MFU by 30‑60% and cut deployment costs.
This article explores why large‑model Agent technology is essential for AIOps, explains single‑ and multi‑Agent architectures, memory and tool integration, and demonstrates practical applications such as anomaly detection, fault diagnosis, automated remediation, ChatOps, and future directions for intelligent, autonomous operations.
The article examines the high construction cost and lack of unified standards in knowledge graphs, explains why large language models cannot fully solve core issues such as hallucination and multi‑hop reasoning, and argues that a new, unified semantic framework integrating large models is essential for future progress.
Huawei Cloud’s Pangu weather large model leverages a 3‑D neural network and AI‑driven temporal aggregation to deliver seven‑day global forecasts up to ten thousand times faster than traditional methods, improving accuracy and lead time over established ECMWF systems.
The article explains practical techniques for choosing and constructing fine‑tuning data for large language models, covering data diversity through similarity‑based clustering, semi‑supervised filtering with binary classifiers, and uncertainty‑driven sampling using perplexity or reward models to build an efficient, low‑cost pipeline.
The article demonstrates how Tencent's Hunyuan large model can generate Python scripts to automatically compare Excel‑based reports, highlight differences, and handle multiple files, turning a tedious manual regression test into a fast, reliable automated process.
This guide introduces Tencent Hunyuan's upgraded text‑to‑image model, explains its technical innovations, provides detailed prompt engineering advice, showcases example prompts and generated images across various styles, and highlights real‑world applications and performance metrics for developers and creators.
This article explains the challenges of large‑model offline (batch) inference, such as GPU memory limits and distributed scheduling, and shows how Ray’s cloud‑native architecture, model partitioning, and Ray Datasets can be used to build efficient, elastic inference frameworks deployed with KubeRay.
NetEase’s Fuxi team developed the multimodal ‘YuZhi’ model, a large‑scale image‑text dual‑tower system optimized with the EET inference framework, which powers personalized recommendations in NetEase News and Cloud Music, while a partnership with Huawei Ascend AI and MindSpore enables further model acceleration, compression, and the new ‘YuZhi‑Wukong’ model that improves video recommendation metrics by about 5%.
This article introduces the rapid development of multimodal digital humans powered by large AI models, covering their background, current challenges, NeRF‑GAN based modeling methods, multimodal dialogue capabilities, and real‑world application cases such as virtual assistants, tourism guides, and sign‑language avatars.
Alibaba’s DAMO Academy unveiled the low‑carbon M6 multimodal model, a trillion‑parameter AI trained on just 480 V100 GPUs, achieving over 80% energy reduction and 11‑fold speedup compared to prior trillion‑parameter efforts, and already powering e‑commerce and manufacturing design tools.
