This paper presents HPN, Alibaba Cloud's data center network for large language model (LLM) training. Due to the differences between LLMs and general cloud computing (e.g., in terms of traffic patterns and fault tolerance), traditional data center networks are not well-suited for LLM training. LLM training produces a small number of periodic, bursty flows (e.g., 400Gbps) on each host. This characteristic of LLM training predisposes Equal-Cost Multi-Path (ECMP) to hash polarization, causing issues such as uneven traffic distribution. HPN introduces a 2-tier, dual-plane architecture capable of interconnecting 15K GPUs within one Pod, typically accommodated by the traditional 3-tier Clos architecture. Such a new architecture design not only avoids hash polarization but also greatly reduces the search space for path selection. Another challenge in LLM training is that its requirement for GPUs to complete iterations in synchronization makes it more sensitive to singlepoint failure (typically occurring on ToR). HPN proposes a new dual-ToR design to replace the single-ToR in traditional data center networks. HPN has been deployed in our production for more than eight months. We share our experience in designing, and building HPN, as well as the operational lessons of HPN in production.

A nano-form composite of MXenes (Ti₃C₂T_x, T_x = -O, -OH, -F) was synthesized through depositing bismuth-nanoparticle (BiNPs) onto Ti₃C₂T_x sheets. Because of the preventive effect of the two-dimensional layered structure of Ti₃C₂T_x, the nanoparticles of Bi were uniform and well attached on the Ti₃C₂T_x. The obtained BiNPs/Ti₃C₂T_x nano-composite was applied for sensors construction of electrochemical detecting of Pb²⁺ and Cd²⁺ heavy metal ions. The produced BiNPs@Ti₃C₂T_x-based sensor showed high effective surface area and excellent conductivity. Also, the BiNPs were efficient for anodic-stripping voltammetric to detect heavy metal ions. After conditions optimization, the BiNPs@Ti₃C₂T_x nano-sensor could detect Pb²⁺ and Cd²⁺ simultaneously and the detection limits were 10.8 nM for Pb²⁺ and 12.4 nM for Cd²⁺. The BiNPs@Ti₃C₂T_x was promising for detecting heavy metal ions due to their high surface area, fast electron-transfer ability, environmental friendliness, and facial preparation.

The nanosized UiO-66-NH₂metal–organic framework (MOF) material was synthesized and modified by palmitoyl chloride to enhance the dispersibility and restrain the aggregation of MOF particles in the organic phase.

Using Chitosan/PEO as the shell and PCL as the core, chitosan–polyethylene oxide/polycaprolactone nanofibrous mats were prepared successfully by coaxial electrospinning for co-load and sequential co-delivery of two drugs. Herein, lidocaine hydrochloride (Lid), used for pain relief, was added to the shell, and curcumin (Cur), an anti-inflammatory agent, was introduced into the core. Sodium bicarbonate (SB) was also added to the core layer to provide wound microenvironment sensitivity. Under acidic conditions, Lid was released due to the formation of –NH₃ ⁺ by protonation of –NH₂ on the chitosan molecular chains. At the same time, SB reacted with hydrogen ions to generate CO₂, and many holes were generated on the surface of the fibers, providing more discharge paths for Cur release. Additionally, both Lid in the shell layer and Cur in the core layer exhibited acidic pH (∼5.4)-responsive release profiles. Moreover, a rapid release of Lid and a sustained release of Cur were observed to provide the immediate effects of analgesia and long-term antibacterial activity in the process of wound healing. Furthermore, after 48 h incubation, the mats showed continuous and excellent antibacterial performance against E. coli and S. aureus. The results of blood coagulation showed that the mats could achieve rapid hemostasis in the early stage of wound formation. Hemolytic and cytotoxicity evaluation also revealed that the mats had excellent hemocompatibility and cytocompatibility. Therefore, this study has made invaluable contributions to the design of a dual-drug-loaded dressing with microenvironment-responsive and sequential release properties towards wound care.