2044 results about "Memory management" patented technology

A platform for coordinating networks of specialized AI agents that enables secure collaboration through token-based communication and real-time result streaming. The system features a central orchestration engine managing interactions between domain-specific expert agents, with memory management and optional encryption for secure data handling. The platform uses efficient communication protocols for knowledge compression and faster reasoning, while a standardized agent interface system handles security, privacy, and policy requirements. It scales across distributed computing environments to enable complex collaborative tasks like personalized content creation, materials discovery, and drug development while optimizing resource usage and maintaining data privacy.

Key-value (KV) cache paging schemes can improve memory management for KV caches by storing a KV cache page having key tensors and value tensors for a fixed number of tokens in a fixed-sized block in the KV cache of a worker. To further improve memory management, the schemes can be modified to implement dynamic variable quantization. Quantization level of a KV cache page can be set based on a runtime importance score of the KV cache page. In addition, the quantization level of the KV cache page can be set based on the system load. The end result is a scheme that can achieve a high compression ratio of KV cache pages in the KV cache. Fitting more KV cache pages in the KV cache can lead to higher inference throughput, higher system-level user capacity, and higher end-to-end service availability.

A dynamic generative-AI enhanced leadership coaching platform includes an agent, an authentication module, an event processor, a user interface module, a router, a generative AI module having a prompt compiler module, a post-conversation analysis module, a prompt template library, an API interface module, a memory management module having a vector-search module, a vectorized database module configured to store vectorized data records and be accessible to the memory management module, a vectorization module configured to vectorize data, a knowledge base database module configured to catalogue vectorized data, and a user profile stored in the user database including a dynamic data point, a personalized conversation starter, and a user permission set.

The invention provides an AI agent memory management method and system based on cold and hot stratification, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining and preprocessing the historical information of an AI agent, and carrying out the vectorization storage of the historical information; calculating a score based on the access frequency and the semantic association; constructing a dual-threshold triggering mechanism to realize data cold and hot layered migration; establishing index mapping and dynamically monitoring data popularity for returning; and a storage water level dynamic adjustment migration strategy is monitored in real time. According to the invention, the problem of low memory management efficiency of the AI agent is solved, efficient access of memory content is realized, and the storage cost is reduced.

The invention provides a large language model intelligent interaction method and system with memory management capability in the technical field of artificial intelligence. The method comprises the following steps: S1, creating a work memory module, a long-term memory module, a memory fusion module and an intelligent interaction module; s2, acquiring external knowledge and integrating the external knowledge into a long-term memory module; s3, the intelligent interaction module obtains an intention sequence based on a user query statement and sends the intention sequence to a work memory module; s4, the memory fusion module retrieves context information from the long-term memory module based on the user query statement and sends the context information to the working memory module; s5, the work memory module constructs a work memory sending intelligent interaction module based on the intention sequence, the context information and the system instruction; s6, the intelligent interaction module outputs a query result based on the working memory; s7, a memory fusion module extracts semantic memory and plot memory from user query statements and query results and stores the semantic memory and plot memory in a long-term memory module. The method has the advantage that the man-machine interaction performance is greatly improved.

A memory management method and apparatus, and a computing device. The method is executed by a GMEM system. When receiving a physical memory range, a memory management unit operation function, and an accelerator device address of an accelerator device, the GMEM system configures a memory management unit operation function interface for the accelerator device, and a device where the GMEM system is located is enabled to use a driver of the accelerator device to manage a memory management unit and a physical memory of the accelerator device by means of the memory management unit operation function interface for the accelerator device, so as to read and write data. In the present application, the GMEM system extends a native memory management interface of an operating system into a heterogeneous memory management interface, supporting programming based on a unified virtual address space, and memory management unit operation function interfaces are configured for different types of accelerator devices, so that the device where the GMEM system is located can support the use of different types of accelerator devices, without the need to consider the compatibility of memory interfaces of different types of accelerator devices.

The invention belongs to the technical field of front-end development, and particularly relates to a React framework-based tree component optimization method fusing virtualization rendering and increment rendering. Comprising the steps of S1, data preprocessing; S2, dynamic virtualization rendering: dynamically calculating a node index range in a visible area according to a real-time scrolling position of a user viewport, only rendering nodes in the range, and realizing seamless lazy loading of data through scrolling event monitoring; s3, intelligent incremental updating: identifying changed nodes through a state change comparison algorithm, and executing local re-rendering only for the changed nodes based on a component life cycle control mechanism of a React framework; and S4, lazy loading cooperative control. The method aims at solving the problems in the aspects of performance, memory management, user experience and the like when a tree component in a React ecological system processes large-scale data, and is widely applied to various Web application program development scenes needing to display and operate a large-scale tree data structure.

The invention relates to the technical field of audio signal processing, and discloses a high-sampling-rate audio analysis optimization method and system, a storage medium and equipment, and the method comprises the steps: upgrading an FFTW (Fast Fourier Transform) library, and building an optimized memory management mechanism; carrying out adaptive framing processing on the input audio data, and optimizing the data access efficiency by adopting a multi-stage cache system; the calculation precision of the FFT fast Fourier transform is dynamically adjusted, and the calculation of the FFT fast Fourier transform is optimized by pre-calculating and caching an FFTW plan; performing clock synchronization optimization and jitter elimination on the COAX digital audio signal; establishing a multi-dimensional result cache matrix and an intelligent multiplexing decision tree, and reducing repeated operation through intelligent cache prediction and similarity calculation; system performance indexes are monitored in real time, adaptive parameter adjustment is realized based on an optimization objective function, audio quality and system stability are monitored in combination with subjective and objective evaluation mechanisms, repeated operation is remarkably reduced, and calculation resource allocation is optimized on the premise of ensuring sound quality.

The invention relates to an implementation method and device for an EtherCAT master station system based on a non-operating system, and the method comprises the steps: achieving the adaptation of an SOEM protocol stack to a hardware abstraction layer of an EtherCAT master station through the byte order processing based on a bit mask and a staged shift strategy and the EtherCAT frame receiving and transmitting control based on annular buffer area index management; on the basis of non-real-time task priority scheduling processing of a hardware timer, main circulation driven by a single-thread event and overtime monitoring of a blocking communication synchronization mechanism in combination with the timer, adaptation of an SOEM protocol stack to an EtherCAT master station operating system abstraction layer is achieved; through network scanning and communication parameter configuration, a memory management mechanism based on CoE standard mapping and a periodic position mode of a slave station, adaptation of an EtherCAT master station application layer is realized. According to the invention, through optimization and adaptation of the SOEM protocol stack, operation-system-free and microsecond-level communication jitter of the EtherCAT master station system is realized.

The invention discloses a data enhancement method and system based on multi-agent self-evolution and hybrid evaluation, and relates to the technical field of artificial intelligence, and the method comprises the steps: employing a teacher model as a multi-agent cooperation system, and carrying out the interaction of a reasoning agent, an evaluation agent, a reflection agent and a memory management agent through a reasoning agent, an evaluation agent, a reflection agent and a memory management agent; iteratively generating a high-quality reasoning data set with a traceable reasoning path and a self-verification label, and performing multi-task supervision fine tuning on a learning model; a diversity sampling strategy based on reinforcement learning is adopted to generate multiple groups of outputs, and a weak point data set is screened by using a consistency score; and for the weak point data set, the multi-view instruction rewriting agent performs diversity rewriting on the instruction and then returns to perform deep reasoning distillation to generate new enhanced data and combine the new enhanced data to high-quality reasoning data. According to the method, multiple agents are arranged, so that the model is iteratively synthesized, evaluated, reflected and modified, the obtained enhanced data can be directly used, a manual auditing step is omitted, and the synthesis efficiency is improved.

The invention discloses an intelligent monitoring method for a parallel DC power supply system based on an edge gateway, and the method comprises the following steps: S1, collecting and processing real-time data, and generating a health baseline; s2, comparing the real-time data with the health baseline, and automatically identifying an abnormal state; s3, generating a fault fingerprint based on a fault fingerprint generation algorithm, and storing the fault fingerprint in a fingerprint database to form a fault file; s4, generating an intention map by using a situation modeling algorithm, dynamically predicting a load demand based on an intention prediction algorithm, and performing adjustment according to a load priority; s5, generating a dynamic power supply parameter adjustment scheme through an evolution rule engine, and performing real-time optimization by adopting a self-evolution algorithm; s6, a multi-time window adaptive memory management module based on an edge gateway performs hierarchical storage and management on real-time data according to time windows; and S7, analyzing the real-time monitoring data based on a closed-loop feedback control system. According to the invention, the edge gateway and the adaptive algorithm are utilized to realize real-time monitoring and optimization of the parallel DC power supply system.

A method includes receiving a machine learning (ML) network model in high-level code; generating an internal representation (IR), the IR is mapped to components in a multi-processing tile device; determining whether a first processing tile with a first on-chip memory (OCM) has a same dimension for an input / output tensor data as a second processing tile with a second OCM performing a same primitive function based on the IR; allocating a same memory address range within the first and the second OCM for the same primitive function if the first processing tile has the same dimension for the input / output tensor data as the second processing tile for the same primitive function; linking the memory address range of the first OCM to the memory address range of the second OCM to form a grouped memory space within the first and the second OCM respectively; and compiling low-level instructions based on the linking.

The invention discloses a large model training fault recovery method based on distributed memory management, and the method comprises the steps: obtaining a model state at a current moment when judging that a check point storage service needs to be carried out in a current training round, serializing the obtained model state into a storable format, and generating check point data; then transmitting the check point data to a host side memory from a device side memory in the node; after the transmission is finished, continuing to train the task, and storing asynchronous check points at the same time; the asynchronous check point storage comprises data distributed memory backup and data persistence to a disk; and when the training task goes wrong, task recovery is carried out through check point data. When the fault is recovered, the backup data is preferentially read from the memory, and the delay caused by the I / O bottleneck of the disk can be obviously reduced, so that the training pause or the efficiency reduction caused by the operation of storing or loading the check point data is reduced, the efficient utilization of the GPU or other computing resources is kept, and the continuity and the stability of large-scale model training are improved.

The invention discloses a self-adaptive memory enhancement imitation learning method for a robot control task, which comprises the following steps of: performing standardization processing and time sequence feature extraction on original multi-modal data, and constructing hierarchical feature representation; establishing a layered structure including short-term memory, working memory and long-term memory, and evaluating memory importance and dynamically updating memory content based on use frequency, task correlation and novelty; retrieving related memories according to the current state and carrying out multi-modal feature fusion; a shaping reward is calculated based on neighbor sample search; and optimizing strategy network parameters through a multi-loss function combination. Through adaptive memory management and a multi-modal feature fusion mechanism, the learning efficiency and execution performance of the robot control system are improved.

According to the system memory management method and system, the electronic equipment and the storage medium provided by the invention, the virtual address information in the access information is extracted, and the virtual address information comprises the target virtual address; under the condition that the target virtual address is determined to be in a first preset virtual address range according to the virtual address information, determining an initial physical address corresponding to an initial virtual address in the first preset virtual address range according to a preset address mapping table; performing physical address calculation processing according to the target virtual address, the initial virtual address and the initial physical address to obtain a target physical address corresponding to the target virtual address; and accessing a system memory based on the target physical address and performing memory management on the system memory. By quickly returning the conversion relation from the virtual address to the physical address in the specified range, the conversion efficiency from the virtual address to the physical address can be improved, a conversion page table does not need to be configured according to MMU granularity, waste of storage space can be reduced, and the address conversion efficiency can be improved.

The invention discloses a user-noninductive cold and hot sensing remote memory exchange system, which belongs to the field of computer storage, and is characterized in that cluster nodes are divided into extensible remote memory service nodes and computing client nodes according to roles; the remote memory node comprises a remote memory exchange server and a memory resource registration module; the computing client node comprises a FrontSwap-based remote memory exchange client, and a remote memory exchange server side is registered as a memory resource which is non-inductive to a user; furthermore, popularity statistics based on a popularity histogram is realized in a kernel mode; the memory exchange behavior is guided through the dynamic decision-making module based on online reinforcement learning, the memory management efficiency and the system performance are remarkably improved, and efficient and non-inductive memory expansion service is provided for user application.

The embodiment of the invention provides a solid-state memory and a memory management method, and the solid-state memory comprises a computing interconnection interface, a nonvolatile memory interface controller, a cache controller, a dynamic random access memory, a flash memory controller and a flash memory medium, the dynamic random access memory comprises a controller memory buffer area and a plurality of persistent memory areas, and instructions and data can be separated through the controller memory buffer area and the persistent memory areas to realize effective isolation of the data and the instructions.

The invention discloses a memory management method and device and computing equipment. The method is executed by a GMEM system, after the GMEM system receives a physical memory range, a memory management unit operation function and an accelerator device address of the accelerator device, a memory management unit operation function interface is configured for the accelerator device, and the device where the GMEM system is located is enabled to operate the function interface through the accelerator device memory management unit. And managing a memory management unit and a physical memory of the accelerator equipment by utilizing a driver of the accelerator equipment so as to read and write data. According to the GMEM system, a native memory management interface of an operating system is expanded into a heterogeneous memory management interface, programming of a unified virtual address space is supported, and memory management unit operation function interfaces are configured for various types of accelerator devices, so that the device where the GMEM system is located can support the use of the various types of accelerator devices, and the operation efficiency of the accelerator devices is improved. And the compatibility problem of memory interfaces of various types of accelerator equipment does not need to be considered.

The invention provides a cold start optimization system and a cold start optimization method for a Python Serverless function. The system comprises a dependency analysis module, a module pre-compiling system, a hierarchical module loader, an inert import mechanism, a code cache optimizer and a self-adaptive optimization system. The dependency analysis module analyzes code dependency in a function deployment stage; the module pre-compiling system pre-compiles the source code into an optimized byte code; the hierarchical module loader loads modules of different hierarchies according to priorities; actual loading of an inert import mechanism delay module; the code cache optimizer caches a hotspot function and a module state; and the self-adaptive optimization system dynamically adjusts preloading and caching strategies. The system further comprises a JIT compiling intensifier, a state storage and quick recovery module, a memory management optimizer and a security mechanism. The method has the beneficial effects that the function cold start time is remarkably shortened, the function execution efficiency is improved, the resource utilization is optimized, and meanwhile, the safety and the expandability are ensured.

Systems and methods for efficient memory management during ray tracing are described. A ray tracing system assigns a memory stack to a ray. The ray, when intersection tested against objects of a node, accesses data that is stored in the memory stack. When data is to be consumed from the memory stack by the ray, the ray tracing system uses a memory pointer associated with the ray to locate the requested data. When data is to be stored to the memory stack, the memory allocation circuitry stores data in a free memory block and uses a linked list to link the memory block with other memory blocks storing additional data for the ray.

The invention provides an agent memory management method and device based on asynchronous processing and a storage medium. The method comprises the following steps: after an interaction event is written into a first storage area, issuing a processing event associated with an event identifier to an event queue, and reading the corresponding interaction event from the first storage area according to the event identifier; when the interaction event passes the importance judgment, converting the interaction event into structured memory data; determining second storage areas based on the type identification field of the structured memory data, and respectively writing the structured memory data into at least one second storage area; and analyzing the query request to obtain a query condition, determining at least one target second storage area based on the query condition, reading structured memory data related to the query request from the target second storage area, performing aggregation processing, and outputting the aggregated memory data to the agent to generate a response result. According to the method, the interaction response delay can be reduced, the model calling cost is reduced, and the memory processing auditing performance is improved.

The invention discloses a memory management method, electronic equipment, a storage medium and a computer program product, and relates to the technical field of computers. The cold and hot marks are set for the memory pages according to the cold and hot scores, memory data of different cold and hot grades can be distinguished, whether data transfer is conducted on the memory data of the memory pages or not is judged by comparing with the types of the storage media where the memory pages are located, and energy consumption can be reduced by transferring the data to the storage media matched with the cold and hot marks; whether the current operation mode of the target storage medium supports data transfer or not is judged before data transfer, the target storage medium is adjusted to be in the operation mode supporting data transfer, all the storage media do not need to be kept in the operation mode supporting data transfer all the time, the lowest energy consumption of the operation mode of the storage medium can be kept, and the operation efficiency is improved. The energy consumption can be reduced by combining the dynamic adjustment of the data position according to the load and the static power consumption management.

The invention discloses a server memory management system and a cluster system, and relates to the technical field of server memory management, the system identifies memory fault information of a server based on a fault reminding signal under the condition that a fault signal is received, and the current load state is determined based on operation data of the server. And matching the corresponding target repair strategy to perform fault repair on the server memory based on the target repair strategy. Therefore, the system can dynamically match different fault recovery strategies according to different fault scenes, so that the technical problem of poor flexibility of a memory fault recovery method in related technologies is solved, and the technical effects of improving the flexibility of memory fault recovery and the service operation stability of a server are achieved.

The invention discloses an intelligent dialogue system for time sequence memory management and task distribution, which is based on CrewAI multi-agent collaboration, combines time sequence memory updating and multi-agent collaboration work, introduces an independent memory unit in order to solve the problem of low memory efficiency management, and uses an undirected graph structure to store learned information. Storing each piece of information as a node in the graph; meanwhile, memory is increased, modified and forgotten according to a time sequence, so that the problem that important information is forgotten is relieved to a certain extent; in order to solve the problem of insufficient task interaction, a multi-agent cooperation mechanism is introduced, user demand understanding, task execution and task verification are respectively submitted to three different agents for independent work, and through independent work and complete information exchange among the three agents, information synchronization and state sharing among modules are realized, information islands are reduced, and the task interaction efficiency is improved. And decision accuracy is improved.

The invention belongs to the field of machine learning, and discloses an FPGA (Field Programmable Gate Array) superposition processor acceleration system and method based on state space duality, which comprises a sparse predefined data acquirer, a reconfigurable systolic array, a partial sum cache, an element-by-element operation cache, a function calculation module, an on-chip memory management module and the like. Zero elements are eliminated through the sparse predefined data acquirer, redundant data are reduced, and redundant calculation is remarkably reduced; the reconfigurable systolic array flexibly supports multiple calculation modes, and the utilization rate of hardware resources is increased; the part and the cache realize cross-cycle accumulation and element-by-element operation cache integration results, the function calculation module completes nonlinear operation, and the on-chip memory management module optimizes result cache, so that on-chip operation of SSD calculation is ensured, off-chip memory access is remarkably reduced, and reasoning efficiency and energy efficiency are improved; by adopting the system, the memory occupation is effectively reduced, the element-by-element calculation efficiency is improved, and the sparse calculation redundancy is reduced, so that the reasoning process of the Mamba2 model is remarkably accelerated.

The invention discloses a hot page threshold adjusting method and device, a medium, a product and a memory system, and relates to the technical field of memory management.The current load index is calculated by introducing multiple load parameters of the memory system, the running state of the system is dynamically reflected, preset hyper-parameters are matched according to the load index, and the hot page threshold is adjusted according to the hyper-parameters. Comprising a first factor influencing the bandwidth and a second factor inhibiting the ping-pong phenomenon, and according to the hyper-parameters, a hot page threshold value is calculated in a self-adaptive mode, so that dynamic classification of the hot page and the cold page is achieved. Compared with a mode of determining the hot page and the cold page by adopting a static threshold, the method has the advantages that the change of the load of the memory system can be responded in real time, the phenomena of resource waste and repeated page migration caused by a pseudo hot page are inhibited, the technical problems of hot page misjudgment and frequent page migration caused by improper hot page threshold setting are solved, and the page migration efficiency is improved. The technical effects of improving the access efficiency and improving the flexibility of memory management and the overall performance of the system are achieved.

The embodiment of the invention discloses an intelligent dialogue method and electronic equipment, and the method comprises the steps: receiving dialogue content inputted by a user in a current dialogue round in a dialogue process of performing dialogue with the user through a large-scale language model (LLM); requesting to obtain the memory content of the current session from a memory management module; wherein the memory management module comprises a first memory module and a second memory module, and the first memory module is used for storing memory abstract content generated according to historical dialogue content; the second memory module is used for storing a newly generated dialogue content original text in the current dialogue; according to the memory abstract content currently stored in the first memory module and the dialogue content original text stored in the second memory module, prompt information used for guiding an LLM model to execute a content generation task is generated. Through the embodiment of the invention, the user request can be quickly responded, and the memory integrity is realized, so that the accuracy of the reply content is guaranteed.

The invention provides a memory resource access method and device, an address translation method and device, electronic equipment and a storage medium. According to the embodiment of the invention, a virtual machine management component can be used for firstly obtaining a direct memory access request of external equipment to a memory resource mounted under a virtual machine; and searching a target physical address corresponding to the target virtual address of the target memory resource from the data processing unit, so that the external equipment accesses the target memory resource by accessing the target physical address. Wherein a physical address translated by the virtual machine management component for the virtual address of the memory resource in advance is stored in the data processing unit, and the data processing unit performs data communication with the virtual machine management component through a communication drive simulated in the virtual machine management component. By adopting the scheme, the congestion of the shared memory resource (IOTLB) of the memory management unit possibly caused by excessive address query can be avoided, the consumption of the shared memory resource is reduced, and the utilization rate of the shared memory resource is improved.

The present disclosure relates to data block retransmission in wireless communications, and particularly in wireless communications having a relatively large propagation delay. Some embodiments relate to memory management and may involve a maximum time period during which a data block that is to be communicated in a wireless communication network may remain in memory space at a user equipment (UE) to support a data block retransmission procedure. UE reporting of current occupancy status of the memory space is also or instead supported in some embodiments. Selective transmission or retransmission may be provided, and may involve reception occasions at which a UE may receive a data block retransmission or a transmission of a different data block.

The invention discloses a memory management method, a program product, equipment and a medium, and relates to the technical field of computers.The memory management method comprises the steps that memory configuration information in the system kernel starting process is obtained, and a reserved memory range is determined based on the memory configuration information; segmenting the reserved memory based on the hot plug framework of the memory and the reserved memory range; allocating corresponding target metadata to each reserved memory segment to obtain a configured physical memory; the target metadata is metadata representing a data structure of the physical memory page; establishing a target mapping array based on the memory configuration information; the target mapping array is used for recording a mapping relation between a physical address and a logic page number in the configured physical memory; responding to the target memory access request by utilizing the target mapping array, and completing corresponding memory access operation; the target memory access request is an access request initiated for the reserved memory, so that the universality of the reserved memory is improved, the overall manageability of the system is enhanced, and the complexity and maintenance burden of the system are reduced.