188 results about "Adapter (computing)" patented technology

The invention discloses a computing system and a control method, device, equipment, medium and product thereof, and relates to the technical field of computers.A host device, a first switch and an acceleration computing device are divided, the first switch comprises a plurality of interconnected switching controllers, and uplink ports of the switching controllers are connected with the host device; the downlink port is connected with the acceleration computing device, the binding relation of the physical link layer between the mainboard and the acceleration card is removed, the port configuration and the resource allocation path of the exchange controller can be flexibly adjusted, and the acceleration computing resources can be allocated according to needs. The first board card provided with the first slot and the cable adapter plate are arranged in the acceleration computing device, so that the acceleration computing device with high-density deployment is realized. Therefore, the method solves the problem that the expansion capability of acceleration computing resources of a multi-card system reaches a bottleneck, and lays a foundation for expanding large-scale computing power by a single host and constructing a multi-host shared large-scale computing resource pool.

The invention provides a component upgrading method and device in a K8s cluster and electronic equipment, and relates to the technical field of cloud computing, and the method comprises the steps: receiving a target message body set through an adapter component under the condition that a component in a first container group POD in the K8s cluster is upgraded; obtaining a first message associated with the first POD from the target message body set through the adapter component; processing and packaging the first message through the adapter component according to the role to which the first POD belongs to obtain a first message body; and updating the target message body set through the adapter component, and sending the updated target message body set to the K8s cluster. According to the embodiment of the invention, when the component in the first POD is in the hot upgrade state, if the first POD is restarted or switched, the state of the first POD in the information acquired by the K8s cluster is not changed, so that the related processing of the K8s cluster on the service cannot be triggered, and the service cannot be disconnected.

The invention provides a multi-LoRA large language model deployment system based on a cloud computing platform. A cloud computing AI platform layer trains a LoRA adapter matched with a basic large language model for a reasoning process; the multi-LoRA dynamic loading layer dynamically switches LoRA adapters needing to be mounted according to the request parameters, GPU optimization configuration is carried out according to service priorities in the request parameters, a basic large language model is loaded, and a plurality of LoRA adapters stacked in a sparse matrix form are mounted; and the resource scheduling optimization layer responds to the resource scheduling application, and outputs the request parameters subjected to priority management and hardware sensing optimization to the LoRA dynamic loading layer based on the container arrangement platform. According to the method, LoRA and container arrangement are deeply fused, and an automatic assembly line of training and reasoning is achieved. A video memory sharing mechanism enables a plurality of service scenes to share the same basic large language model, a LoRA adapter is loaded as required, and video memory occupation is greatly reduced. Full-life-cycle management from data preparation to model service is supported, and the large model deployment cost is remarkably reduced.

The invention discloses a large model fine tuning training method, device and equipment, a storage medium and a product, and relates to the technical field of network intelligence, and the method comprises the steps: inputting a label data set into a large model through end side equipment for forward propagation calculation, and transmitting an output activation value between each Trm layer to auxiliary computing power node equipment; and the auxiliary computing power node equipment fuses the output activation value with the output of each layer in the bypass network adapter to execute forward propagation calculation, and the end side equipment fuses the target output with the output of the last Trm layer to obtain model output. And determining a loss value according to the model output and the local tag data set, so that the auxiliary computing power node equipment performs back propagation parameter updating according to the loss value. Therefore, a backbone network passing through the end-side device is avoided, the total memory overhead in the training process is reduced, the memory pressure of the end-side device is effectively migrated to the auxiliary computing power node, and the end-side device with the limited memory efficiently executes a large-model fine tuning task.

A non-transitory storage system is constructed based on the converged architecture system and includes: a plurality computing nodes in a computing resource pool, an adapter board, and a plurality of solid-state drives configured to establish a communication connection with the plurality of computing nodes through the adapter board to expand a storage resource pool. Through the foregoing design, the storage resource pool is expanded based on the third-generation converged architecture system and a bus interconnection technology, to achieve remote expansion of storage. In addition, the plurality of computing nodes share decoupled storage resources, thereby solving the problems that a storage capacity of conventional Just a Bunch of Flash (JBOF) may only be expanded to a limited extent and remote decoupled access to storage resources by a plurality of hosts is not capable to be achieved.

A software defined distributed control system (SDCS) in a process plant includes an application layer that includes a plurality of containers instantiated in a data cluster. Each of the containers is an isolated execution environment executing within the local operating system of a respective computing node. The containers cooperate to facilitate execution of a control strategy in the SDCS, and includes a hyper converged infrastructure (HCI) operating across the data cluster, which HCI is configured to communicate with the application layer via an adapter service. The HCI includes software-defined (SD) compute resources, SD storage resources, SD networking resources, and an orchestrator service. The orchestrator service is programmed to configure a first container to include a service executing within the first container. It also assigns the first container to execute on an available hardware resource to control a plurality of field devices operating in the process plant.

The invention provides an IP network acquisition and control plug-in implementation method and device based on a Matrix command adapter, and relates to the technical field of network communication, and the method comprises the steps: analyzing and converting a control instruction of heterogeneous network equipment through the Matrix command adapter, and generating a standardized intermediate instruction set; deploying a resource virtualization pool in the IP network acquisition and control plug-in, and dynamically allocating network bandwidth, computing resources and equipment interfaces; and matching the middle instruction set with the execution capability of the target equipment in real time, and generating an executable equipment control instruction. According to the method, heterogeneous network equipment instructions can be flexibly analyzed and converted through the dynamic protocol adaptation layer, new equipment and new protocols can be quickly adapted through template matching and self-defined script extension, the compatibility and expansibility of network management are improved, and the resource virtualization pool is combined with the prediction model and the dynamic priority queue, so that the network management efficiency is improved. The resources are intelligently allocated according to the real-time load and historical data of the equipment, the resource utilization rate is improved, operation of core service equipment is guaranteed, and the overall performance of the network is improved.

The embodiment of the invention relates to the field of computing power scheduling, and discloses an intelligent inference system and method, equipment and a medium. The system includes a base model for performing an inference process of shared parameters; a plurality of adapter models, each of which is used for performing an inference process of an inference task; the base model and the adapter model are separately and independently deployed, and parallel inference of the multiple adapter models on one base model is supported; and the intelligent router is used for receiving and analyzing the inference task, selecting the corresponding adapter model according to a preset rule, and loading the adapter model to the corresponding inference node. By introducing the adapter design of the LoRA model and the intelligent router mechanism, the efficient deployment and inference of the multi-expert model in the low video memory environment are realized, the use of the video memory can be obviously reduced, and the inference response speed is improved at the same time. The technical problem of low expert model inference efficiency can be at least solved.

The invention provides an open accelerator infrastructure server which can be applied to the technical field of servers. The server comprises an acceleration node comprising a universal substrate, a plurality of accelerator modules and at least one switch chip, the plurality of accelerator modules and the at least one switch chip being arranged on the universal substrate; the computing node comprises a server main board and a plurality of computing units arranged on the server main board, each computing unit is provided with a network interface card, and the server main board and the universal substrate are arranged in parallel; the front panel is provided with M structure pluggable modules and N network interface cards, the front panel is used for interconnection between the acceleration node and at least one computing node and interconnection between the acceleration node and an acceleration node outside the open accelerator infrastructure server, and the front panel is perpendicular to the universal substrate.

A misinsertion detection method for an external device, and a computing device, relating to the technical field of servers, for use in ensuring the correct connection between a docking station and a computing device, so that the computing device can normally use a hardware device to be expanded that is inserted on the docking station. The computing device comprises a controller, a plurality of adapter connectors, and at least two docking stations; each docking station comprises at least two expansion connectors; one of the plurality of adapter connectors is connected to one expansion connector; the plurality of adapter connectors are separately connected to the controller; and the controller is configured to: determine a current connection relationship of each docking station, the connection relationship comprising a connection relationship between each expansion connector of the docking station and the corresponding adapter connector; and output alarm information when the current connection relationship of the docking station is different from a preset connection relationship of the docking station, the alarm information being used for indicating that the current connection relationship of the docking station is incorrect.

The invention relates to the technical field of block chain smart contracts, and provides a smart contract automatic execution system and method based on a block chain technology. In order to solve the problems that in the prior art, out-of-chain event verification is fragile, cross-contract atomicity is lacked, and resource management and control are insufficient, the system comprises a multi-mode under-chain event verification module, and a tamper-proof data channel is established through an Internet of Things hardware anchor point and an anti-collusion oracle; the autonomous execution engine is integrated with a resource awareness scheduler and an out-of-chain computing adapter; the transaction atomicity guarantee module is used for realizing cross-contract rollback by adopting a global snapshot tree and an intelligent compensation contract; the governance arbitration interface supports anonymous dispute resolution. The method comprises the steps of out-of-chain event multi-level verification, voucher execution queue management and sandbox resource isolation control. Credible collaboration inside and outside the chain is realized, the execution reliability is improved by more than 90%, and the contract development cost is reduced by 40%.

The invention provides a personalized federated learning method based on personalized enhancement and sharing self-adaption, which comprises the following steps: enabling a sharing module in a Transform model to participate in federated aggregation, promoting knowledge migration of each client, storing a personalized module in a local client, and improving the adaptability to heterogeneous data; a Lora low-rank adapter is introduced into a personalized module, only Lora parameters are uploaded, feature expression homogenization caused by global aggregation is avoided, initial global shared module parameters and local parameters of a shared module are subjected to self-adaptive fusion, semantic offset caused by model decoupling is effectively relieved, and the robustness of the system is improved. Therefore, the generalization performance is improved, and the stability and adaptability of a local model are kept; dynamic incremental clustering is performed on each client based on Lora parameters, and shared feature representation with cross-client migration value is mined in a personalized module, so that the calculation overhead is reduced, and the personalized knowledge sharing and migration capability among similar clients is enhanced.

Techniques are described with respect to managing a distributed device message in a computing infrastructure. Such techniques are enabled through a universal interface apparatus including a plurality of serial interface adapter boards and a system-on-a-chip microcontroller. The universal interface apparatus provides a universal gateway solution between one or more component interfaces associated with a certain premises or environment and a remote system. An associated method includes deriving core message content from a distributed device message originating from a source component in a computing infrastructure, converting the derived core message content to open standard file format message content, propagating the open standard file format message content to a virtualized management system, and receiving an open standard file format message response from the virtualized management system.

The invention relates to the technical field of electric power design, and discloses an electric power design agent intelligent processing system based on an AI large model, which comprises an interface interaction module, a professional model module, a dialogue management module and a professional algorithm module. The interface interaction module orderly calls idle modules according to the instruction sequence; the professional model module periodically retrieves multi-source data, constructs a professional data set through dynamic denoising, secondary classification processing and verification, generates a fine tuning data set through various knowledge distillation modes, and optimizes an industry large model by adopting a low-rank adapter; the dialogue management module is connected with input and output and instruction generation; and the professional algorithm module dynamically calls five types of professional models and adapts to computing power resources. According to the system, a traditional tool data island is broken through, the problems of general LLM professional shortages and'illusion 'are solved, automatic closed loop from natural language interaction to professional design tasks is achieved, manual dependence and error risks are reduced, and power design efficiency, quality and system stability are improved.

Systems, methods, and apparatuses disclosed herein can enable a computing system to connect to a network. These systems, methods, and apparatuses can connect the computing system to the network through multiple network connections. These multiple network connections represent distinct, physically separate signal pathways to improve security. This physical separation, or isolation, from one another creates distinct boundaries between these multiple network connections, for example, to minimize shared resources these systems, methods, and apparatuses. These distinct boundaries can reduce the vulnerability of these systems, methods, and apparatuses to, for example, attacks that exploit shared sources among these systems, methods, and apparatuses, data leakage within these systems, methods, and apparatuses, and / or unauthorized access to these systems, methods, and apparatuses. Moreover, these distinct boundaries can additionally enhance security by improving fault isolation, enforcing access control, and / or supporting secure communication protocols, among others, within these systems, methods, and apparatuses.

The invention provides a modular data center machine room system, and belongs to the technical field of intelligent machine rooms, the modular data center machine room system comprises a plurality of stackable functional unit modules, and each functional unit module comprises a composite load-bearing frame, an integrated standardized mechanical interface and an electrical fast-assembly interface; the computing unit supports plug and play of heterogeneous computing equipment; the storage unit supports RAID hot plug of the mixed medium; the network unit provides a programmable switching matrix; the power supply unit is internally provided with a multi-mode adapter and is compatible with AC / DC input and lithium battery / super capacitor hybrid energy storage; and the central intelligent management platform monitors the real-time load rate and the energy efficiency ratio of each module, coordinates the resources of each module, detects the interface attribute of a new module and configures a network address when judging that the new module needs to be accessed, executes a topology reconstruction instruction, carries out topology reconstruction and outputs a prompt. Compared with the prior art, the crossing from a fixed framework to elastic ecology is realized.

The subject technology relates to model compatibility for large language models. An apparatus receives a first trained machine learning model having a first adapter layer and generates a second trained machine learning model having a second adapter layer, in which the second model is a transformed version of the first model and both models share a base model. The apparatus initializes the second adapter layer using parameters derived from the first adapter layer and trains the second adapter layer using parameters of the first adapter layer and initialization parameters of the second adapter layer. The apparatus computes a divergence metric between probability distributions of the two models to assess a difference between them. The second adapter layer may be adjusted based on the divergence metric exceeding a threshold. The apparatus deploys the second trained machine learning model in a computing environment based on the divergence metric not exceeding the threshold.

An instruction executing within a computing environment is used to perform at least one operation on one or more controls of a queue to be used in communications between a control program executing within the computing environment and an adapter module of the computing environment. Execution of the instruction includes obtaining an indication of the queue for which at least one operation is to be performed on the one or more controls of the queue. At least one operation is performed on the one or more controls of the queue and a result of executing the instruction is provided.

The embodiment of the invention provides a processing method and device for a network card in a server, equipment and a medium, and relates to the technical field of servers, and the method comprises the steps: obtaining an in-place signal of an interface riser card in a second computing node through a first computing node in response to a power-on signal of the server; if the in-place signal is the low-level signal, acquiring a first control signal of a first logic device in the first computing node and a second control signal of a first high-speed connector in the first computing node through the first computing node; performing priority ranking on the first control signal and the second control signal to obtain a control signal sequence for a shared interface network card in the first computing node; the first computing node obtains the second computing nodes from the computing nodes according to the control signal sequence, connection between the second computing nodes and the shared interface network card is sequentially established according to the generated computing node sequence, the purpose that the computing nodes share the same shared network card is achieved, and the cost of server components is reduced.

A control program executing on a computing device of a computing environment transmits control request information to an adapter module executing within the computing environment. The transmitting uses a control plane transmit queue of a set of queues. The control plane transmit queue identifies a location to store the control request information to be retrieved by the adapter module. The control program receives from the adapter module an indication of another location storing control reply information replying to the control request information. The indication is obtained from a control plane receive queue of the set of queues. The control program retrieves from the another location the control reply information and performs processing based on the control reply information.

The invention relates to the technical field of data security and privacy computing, in particular to an ODRL-based trusted data space data delivery method and system, and the method comprises the following steps: defining data use permission and constraint through an ODRL strategy, verifying a user request by a control plane based on the ODRL strategy, and recording a structured log by a data plane through a protocol adapter factory; the method has the beneficial effects that through layered architecture design, ODRL deep integration, protocol irrelevant adaptation and full-link security auditing, the core problems of an existing data delivery scheme in the aspects of authority management, transmission flexibility, performance and compliance are systematically solved, and a universal and highly reliable technical solution is provided for trusted data space construction; and the data circulation efficiency and security are remarkably improved, and the method has wide industrial application value.

A hardware card that can be implemented in a server system is disclosed. The hardware card may be, for instance, a network interface card coupled to a board such as a motherboard or backplane in a computing system. In various instances, the hardware card reports a temperature to a management controller on the board (e.g., a baseboard management controller) that is adjusted due to thermal degradation of the hardware card. For example, the temperature output by the hardware card may be adjusted based on a comparison of a current temperature profile (including transient temperature of the hardware card) to a predetermined temperature profile stored on the hardware card.

The invention relates to an access control method and device and computer equipment. The method is applied to a remote network interface card of a server, and comprises the following steps: in response to an access permission acquisition request of a target shared storage device sent by a current computing node, acquiring access permission information of the target shared storage device for the current computing node, and sending the access permission information to the current computing node, and indicating the current computing node to control to access the target shared storage device according to the access permission information. By adopting the method, the virtual machines on the plurality of computing nodes can be controlled to perform coordinated access on the shared storage device, so that data acquired by the plurality of virtual machines on the plurality of computing nodes are prevented from conflicting, the data consistency can be ensured, and the fence reading and writing capability is improved.

An identifier, such as a worldwide name, corresponding to a host computing device, or corresponding to a host bus adapter corresponding to the host computing device, may be transmitted to a storage array to configure the storage array to be accessed by the host computing device. The host may request a token from the array. The host may embed the token into a token block of a command message that is transmitted to the storage array requesting access to the array. Before performing an action corresponding to the command message, the storage array may verify that the token is a valid token and thus that the host computing device that transmitted the command message is a host computing device that has been authorized to access the storage array. If the storage array determines that the token is valid the storage array may perform an action corresponding to the command message.

Embodiments of the present disclosure implement hardware-based snoop logic streaming timers for input / output (I / O) communications between a host and a shared network adapter in a computing system. The disclosed embodiments describe a hardware-based snoop logic and snoop logic timer function control of input / output (I / O) processor monitoring a defined memory area to detect a completion queue entry being written to a completion queue, and implementing streaming timers and interpacket timers, based on the completion queue entry and predefined configuration information for an associated host interface connection, to provide streaming data status, interpacket arrival times, and streaming timer expiration for the associated host interface connection.

An assembly kit for a computing system is disclosed. The assembly kit includes a mounting ear configured for attachment to a computer chassis, the computer chassis being configured to be received in a cabinet, the mounting ear having an L-shape formed by a first wall and a second wall, the second wall being perpendicular to the first wall, wherein the mounting ear is configured to receive a fastener configured for attaching the assembly kit to the cabinet; and an adapter releasably coupled to the mounting ear, the adapter having a third wall, a fourth wall that is perpendicular to the third wall, and a fifth wall that is perpendicular to the third wall, the fourth wall and the fifth wall being parallel. A space is formed between the fourth wall and the fifth wall, the space being configured to receive the fastener.

This invention discloses a non-embedded UAV information acquisition device and method. The non-embedded UAV information acquisition device includes a shell assembly, a three-mode adapter structure, a standardized physical interface, a main control module, a communication module, a positioning and sensing module, and a power management module. The internal cavity wall of the shell assembly is fitted with an anti-magnetic shielding layer made of permalloy. The three-mode adapter structure is integrated into the bottom of the shell assembly and includes a magnetic suction component, a suspension component, and a screw fixing interface. The standardized physical interface is located on the surface of the shell assembly and uses a gold-plated gold finger contact structure, which is electrically connected to the main control module. This invention enables deployment without disassembly of the UAV through a non-intrusive design while maintaining the UAV's airworthiness. It utilizes edge computing to alleviate cloud scheduling pressure and reduce task latency, and with independent power supply, achieves long endurance and high reliability, making it widely adaptable to various UAV models and harsh operating environments.

The application belongs to the technical field of collaborative learning, and particularly relates to a parameter efficient collaborative learning method and system suitable for intelligent sensing devices. The application is directed to a collaborative application scene of multiple distributed terminal devices with sensing, computing and communication capabilities. Under the premise of freezing backbone parameters, a shared adapter and a personalized adapter are introduced. Personalized parameters are only updated and saved locally to maintain single-device adaptation capability. Shared parameters realize knowledge sharing through interaction and weighted aggregation of neighboring devices. An adaptive weight matrix is used to fuse shared prediction and personalized prediction, so that global collaboration and personalized adaptation are considered in the distributed sensing scene with significant data distribution difference. Meanwhile, only the lightweight trainable part is updated and exchanged, which significantly reduces the end-side computing burden and communication overhead, and maintains the efficiency and convergence stability of collaborative learning under resource-limited conditions.

Techniques discussed herein relate to monitoring and / or detecting destabilization events of a host machine of a cloud computing environment. At least some of these destabilization events can be associated with a converged network adaptor of the host machine. The converged network adaptor may include a combination of functionality of a host network interface card (NIC) and functionality associated with a smart adaptor (e.g., a smart NIC). The smart adaptor may be configured to use at the host machine. A converged infrastructure management service may monitor operational data associated with the host machine to identify destabilization events that may be associated with the converged network adaptor. One or more remedial actions may be identified based at least in part on identifying these destabilizing events.