Tagged articles
large model training
26 articles · Page 1 of 1
DeepSeek plans to pre-train its next flagship model V5 on Huawei Ascend 950DT chips by late 2026, marking a shift from inference to training for domestic AI hardware; the collaboration includes co-design of two Ascend generations, open-source inference framework DeepInfer supporting both Ascend and NVIDIA GPUs, and deployment practices achieving 155 trillion tokens per day.
AI chipsDeepInferDeepSeek
0 likes · 8 min read

This analysis examines AI super node architecture as the solution to large model training and inference bottlenecks, detailing NVIDIA's Vera Rubin NVL72 platform with 7 chips and 5 rack types, China's domestic super node solutions from Huawei, Sugon, Alibaba, and Baidu, and projects a 341% CAGR reaching 1.1 trillion RMB by 2028.
AI super-nodeInference OptimizationNVIDIA Vera Rubin
0 likes · 10 min read

Huawei unveils Atlas 960E SuperPoD, the industry's first super node using Near Package Optics (NPO) technology, scaling to 4,096 Ascend 960 chips delivering 8 EFLOPS FP8 and 16 EFLOPS FP4 for trillion-parameter model training, with 5,500 Hi-ONE optical engines replacing 48,000 modules to cut 550 kW power and achieve 99.8% availability.
AI super-nodeAscend 960Atlas 960E
0 likes · 7 min read

Biren Technology details a three-tier AI super node architecture scaling to 1024 GPUs via NPO optical interconnect and the BLink 2.0 protocol, built on its Chiplet-based BR20X GPUs with native FP8/FP4 support for trillion-parameter model training.
AI super-nodeBLink 2.0BR20X GPU
0 likes · 6 min read

In large‑model training, a smooth loss decline is only a necessary condition; the article explains which additional logs—learning‑rate, gradient norm, parameter updates, validation metrics, throughput, GPU usage, and checkpoint status—must be monitored and how to diagnose common issues with concrete code examples.
GPU utilizationPyTorchcheckpointing
0 likes · 22 min read

The whitepaper defines supernodes as high‑speed, tightly‑connected compute systems with unified memory addressing, microsecond‑level latency and terabyte‑per‑second bandwidth, outlines their physical, transaction, function and topology layers, demonstrates AI training and inference gains such as 80% communication reduction and 98.4% cluster scaling efficiency, and discusses industry impact, future scaling, standardization and green energy trends.
AI infrastructureSupernodehardware-software co-design
0 likes · 8 min read

SpaceX has replaced JAX with a C‑based training stack that Elon Musk says speeds up large‑model training by an order of magnitude, while simultaneously building the 1‑GW Colossus II supercomputer, listing AI infrastructure as a core business, and offering short‑term compute rentals such as a 180‑day lease to Anthropic.
AI computeAnthropicC language
0 likes · 6 min read


AI WakaMay 15, 2026 · Artificial Intelligence
The article analyses how Cerebras' wafer‑scale AI engine reshapes training and inference by offering massive on‑chip memory and bandwidth, simplifying parallelism, altering software stacks, and presenting trade‑offs against mature GPU clusters for large‑model workloads.
AI hardwareCerebrascompiler stack
0 likes · 17 min read

Twinkle is a lightweight client‑server training framework open‑sourced by ModelScope that abstracts away Ray clusters, data and model parallelism, offers three run modes (torchrun, Ray, HTTP), multi‑tenant LoRA training, dual back‑ends (Transformers and Megatron), and a serverless Training‑as‑a‑Service gateway for enterprise and individual developers.
LoRAModelScopeTaaS
0 likes · 14 min read

The article explains why mastering large‑model training is crucial for professionals, researchers, and enterprises, outlines the end‑to‑end pipeline—from data preparation and pre‑training to instruction fine‑tuning and RLHF alignment—compares training with RAG, and presents a structured learning roadmap.
AI agentsData EngineeringPyTorch
0 likes · 14 min read

The article explains how AI large‑model training demands massive GPU resources and how carefully designed network architectures—such as Clos/Fat‑Tree, Spine‑Leaf, multi‑rail versus single‑rail connections, Dragonfly, and Torus—impact performance, scalability, cost, and reliability, guiding the selection of optimal data‑center networks.
AIGPU clustersHigh-Performance Computing
0 likes · 9 min read

The whitepaper explains that AI model training and inference rely on massive data computation, with model sizes reaching billions of parameters, demanding low‑latency, high‑bandwidth, stable, scalable, and manageable networks; it compares RDMA‑based InfiniBand and RoCE solutions and offers design recommendations for future AI compute clusters.
AIInfiniBandRDMA
0 likes · 10 min read

The article introduces the NorthStar large‑model training framework developed by DeWu, detailing its background challenges, pipeline architecture, rich API support, multi‑GPU training modes, multi‑level embedding storage, hardware selection considerations, and a brief Q&A on data versus model parallelism.
AI frameworkEmbedding Storagelarge model training
0 likes · 9 min read

This article analyzes common GPU efficiency problems in enterprise AI compute platforms—such as low utilization, long fault‑resolution times, and limited performance gains—and presents three practical solutions: dynamic resource allocation, systematic fault‑tolerance, and system‑level tuning, illustrated with real‑world case studies.
AI PlatformGPU utilizationlarge model training
0 likes · 11 min read

The article details Baidu Baige 4.0’s four‑layer AI infrastructure—hardware, cluster components, training‑inference acceleration, and platform tools—highlighting its heterogeneous computing, high‑performance networking, fault‑tolerant communication library, and optimizations that boost large‑model training and inference efficiency.
AI infrastructureheterogeneous computinghigh performance networking
0 likes · 17 min read

Baidu's HPN network solves hash‑collision bottlenecks in large‑model training by combining TOR‑affinity scheduling with Dynamic Load Balancing on self‑developed switches, boosting physical network bandwidth efficiency to about 95%, improving throughput by roughly 10% and adding a further 1.5% training‑speed gain via the BCCL library.
Baidu CloudCollective CommunicationDLB Dynamic Load Balancing
0 likes · 12 min read

This article examines the impact of GPU communication bottlenecks on large‑model training, analyzes hash‑collision issues in high‑performance networks, and presents three practical solutions—including increasing RDMA streams, affinity‑aware scheduling, and dynamic load balancing—to boost effective network bandwidth up to 95%.
RDMAdynamic load balancinghash collision
0 likes · 11 min read

PAI‑TorchAcc, an Alibaba Cloud AI platform accelerator, offers a seamless PyTorch interface that integrates HuggingFace models and employs LazyTensor‑based static graph conversion, multi‑strategy distributed training, and extensive GPU optimizations to dramatically boost throughput for 1B‑175B parameter models, surpassing PyTorch native and Megatron‑LM performance.
AI accelerationAlibaba CloudGPU optimization
0 likes · 13 min read

The article introduces PAI‑ChatLearn, a flexible and high‑performance framework developed by Alibaba Cloud's PAI team that supports full‑pipeline RLHF training for large models, explains the evolution of parallel training strategies, details the framework’s architecture and configuration, and showcases performance results and practical usage examples.
AI frameworkPAI-ChatLearnRLHF
0 likes · 17 min read

The 2023 Open Acceleration Specification AI Server Design Guide details the hardware architecture, OAM module and UBB board specifications, cooling, management, fault diagnosis, and software platform needed to build high‑performance, scalable AI compute clusters for large‑model training.
AI accelerationOAMUBB board
0 likes · 10 min read

This article explains the challenges of training massive AI models and details PaddlePaddle's 4D hybrid parallelism, MoE acceleration, long‑sequence strategies, end‑to‑end performance optimizations, and practical code examples for building and scaling large models efficiently.
AIPaddlePaddleParallelism
0 likes · 12 min read

The article details Baidu's development of a massive high‑performance GPU/IB cluster, its architectural design, the challenges of training trillion‑parameter models, and how the integrated AI stack—spanning hardware, framework, and resource management—overcomes compute, memory, and communication bottlenecks to accelerate large‑model training.
AI infrastructureBaidu AI BaseGPU cluster
0 likes · 17 min read

This article explains how Baidu built a massive, high‑performance GPU/IB cluster, optimized its architecture and software stack, and integrated AI frameworks and resource management to overcome compute, memory, and communication bottlenecks, enabling efficient training of trillion‑parameter large models.
AI infrastructureGPU clusterscloud computing
0 likes · 19 min read

Tencent Cloud's new HCC high‑performance computing cluster triples previous generation performance with 3.2 TB/s server bandwidth, Xingsha servers and NVIDIA H800 GPUs delivering up to 1979 TFlops, while its Xingmai 3.2 T ETH RDMA network, TB‑level storage via COS + GooseFS, and multi‑form access (bare metal, cloud servers, containers, functions) enable efficient large‑model training.
AI computingGPU clusterHigh-Performance Computing
0 likes · 9 min read

This article analyzes Tencent's AngelPTM framework, detailing its ZeRO-Cache strategy, unified storage management, multi‑stream async execution, SSD tiered storage, and performance benchmarks that show up to 95% larger model capacity and over 44% speedup compared to community solutions.
AI infrastructureGPU accelerationZeRO-Cache
0 likes · 12 min read

This article explains how Baidu's intelligent cloud overcomes the compute and storage walls of large‑scale model training by combining hardware design, network topology, and software optimizations such as pipeline, tensor, and expert parallelism, cost‑model‑driven placement, and future‑proof AI infrastructure evolution.
AI infrastructureBaidu CloudGPU optimization
0 likes · 28 min read
