889 results about "Memory management" patented technology

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 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.

A system and method for implementing a Persistent Cognitive Machine (PCMs) that extends beyond the traditional prompt-response paradigm of artificial intelligence are disclosed. A PCM maintains persistent cognitive processes regardless of external interaction, stores and organizes thoughts in a thought cache, retrieves relevant thoughts based on current stimuli, generates new thoughts through reasoning processes, and curates stored thoughts during periods of reduced external interaction. The PCM includes language and reasoning model components, a thought cache, an executive component, and an embedding system. The PCM remains continuously active, remembers previous experiences, learns from these experiences, creates new thought experiences independently, and initiates interactions without waiting for external prompts. The PCM enters sleep-like states during which it curates its thought cache, generalizes experiences, and performs other memory management functions. Applications may include but are not limited to synthetic cognitive colleagues, strategic war gaming platforms, and personal cognitive assistants.

The invention relates to a video memory management method and device, electronic equipment and a storage medium, the method is applied to a GPU drive in a virtual machine, and the method comprises the steps that when it is detected that the remaining amount of a GPU video memory is lower than a set remaining amount threshold value, cold pages in a GPU page table are recognized; wherein the cold page is a page table item which is not accessed in a recent preset time interval; and under the condition that the cold page is a video memory, replacing a physical address corresponding to the virtual address in the cold page to a physical address in a system memory so as to replace video memory data corresponding to the cold page to the system memory. The embodiment of the invention can improve the overall utilization rate of the video memory.

The invention discloses an equipment access method, device and equipment based on address space identification. The method is applied to simplifying an equipment memory management unit and comprises the steps of establishing each single-level range table; acquiring an equipment access request, reading an address space identifier and a virtual address from the equipment access request, and matching a corresponding target range table from each single-level range table based on the address space identifier; and calculating a target physical address according to the target range table and the virtual address, and establishing equipment access based on the target physical address. By replacing a traditional multi-level page table with a single-level structure, the data structure design of address mapping is simplified, the dynamic configuration complexity is reduced, and hardware formal verification is facilitated. Through direct matching of the address space identifier, the traditional steps of analyzing the equipment identity identifier and querying the external table item are omitted, and the external memory access frequency is reduced. By calculating the target physical address and establishing access, the hardware address conversion logic is simplified, the hardware implementation cost and power consumption are reduced, and the data transmission efficiency is improved.

The invention provides an accelerator and a request processing method, which can be applied to the technical field of computers. The accelerator comprises: a computing unit configured to send an access request to a target virtual address to a memory management unit; the memory management unit comprises an address translation buffer module, and the memory management unit is configured to respond to the received access request and search a corresponding physical page number in the address translation buffer module according to the target virtual address so as to determine a physical address; wherein the address translation buffer module comprises a plurality of buffer units, and the address translation buffer module is configured to determine a target buffer unit from the plurality of buffer units according to a virtual base address in a target virtual address; and based on address identification information in the target virtual address, searching a physical page number corresponding to the target virtual address from physical page numbers corresponding to the at least two different virtual base addresses stored in the target buffer unit to determine a physical address.

A system for encoding and compressing images and textures to a bitstream as an executable program of computer instructions operable on pixel space, permitting the Encoder to arbitrarily determine the order, location and method used for reconstructing the components of an image at the decoder. The bitstream is typically further compressed by an Adaptive Entropy Compressor. Bitstream execution reconstructs the image directly to pixel space. Memory management and parallel processing controls may be embedded in the bitstream. The system is optimised for AI processing of images with a universally interpretable construction format that enables machines and humans to understand precisely how the image was created and can be reconstructed. Dual bitstreams may be combined at the decoder for securely embedding customised content (e.g., advertising) at the edge.

The invention relates to a customer service scene-oriented collaboration system and method, and belongs to the technical field of artificial intelligence. In the system, a shared memory storage center stores knowledge entries from a plurality of customer service agents in a predefined data structure; the shared memory storage center is provided with an access control mechanism associated with a visible domain identifier, and an access range is limited during retrieval through a logic partition; the memory management module receives candidate knowledge entries from each customer service agent based on a trigger event, determines visible domain identifiers for the candidate knowledge entries, associates the candidate knowledge entries with the visible domain identifiers, and stores the associated candidate knowledge entries and the visible domain identifiers in the shared memory storage center; the problem processing and decision-making module responds to an access request from any customer service agent and determines a corresponding visible domain identifier according to the identifier of the customer service agent; and executing access based on the visible domain identification and the input question, and retrieving knowledge entries with access from a shared memory storage center to realize cross-agent multiplexing.

A long-term continuous learning method based on task core memory management and consolidation aims to enable a model to sequentially learn from a large number of task sequences, new knowledge is obtained, information of previous learning is reserved, and the method is similar to a human learning mode. The method comprises the following steps: 1) task input and instruction fine tuning; 2) performing difference analysis on the model parameters of the current task and the previous task, identifying a task core memory unit, calculating an adaptive weight based on task prototype similarity, and dynamically updating the memory unit; 3) constructing an experience playback buffer area through a difficult sample selection strategy and a difference sample selection strategy; and 4) utilizing the joint loss function training model to keep the memory of the historical tasks while learning the new tasks. According to the method, the problem of disastrous forgetting in long-term continuous learning is mainly solved, and the performance of the model in a long-term sequence task is remarkably improved.

The invention discloses a memory management method, device and equipment and a computer readable storage medium. The method comprises the following steps: when a mapping request exists, determining a target I / O virtual address corresponding to the mapping request and triggering an application identity identifier of the mapping request; detecting whether the target I / O virtual address is associated with the application identity label or not; and if yes, establishing mapping between the target I / O virtual address and the corresponding memory physical address. According to the method and the device, the memory protection capability of the equipment can be enhanced on a system without IOMMU (Input / Output Management Unit); and the complexity of hardware design can be reduced on a system with an IOMMU, and the hardware cost is reduced.

The invention discloses a memory management method and related equipment, and relates to the technical field of data storage, the method comprises the following steps: based on access information of memory pages of a target operating system, determining a dynamic popularity score of each memory page; on the basis of the current memory pressure level and the dynamic popularity score of the target operating system, executing a corresponding hierarchical data migration strategy so as to migrate different memory pages to the corresponding main Swap storage space or auxiliary Swap storage space; adjusting a compression algorithm applied to the auxiliary Swap storage space based on the current central processing unit utilization rate of the target operating system; and detecting real-time states of the main Swap storage space, the auxiliary Swap storage space and the physical memory, and adjusting a hierarchical data migration strategy and a compression algorithm based on a detection result. Through dynamic popularity perception, hierarchical data migration, adaptive compression regulation and cross-layer collaborative optimization, the memory utilization rate, the exchange efficiency and the overall system stability can be improved under different system loads.

The invention provides a cross-platform video decoding method and device, computing equipment and a computer readable storage medium, and the method is combined with a factory mode to completely package bottom layer differences of all platform types, and an application layer developer does not need to care about the bottom layer platform differences. The optimal implementation can be automatically selected on different engines and platforms through a uniform API (Application Program Interface); during rendering, a single-frame memory management strategy is used, and only one frame which is currently displayed is reserved in a memory; and in the aspect of performance, the advantages of a multi-thread platform and the parallel computing characteristics of the GPU are fully utilized. According to the embodiment of the invention, through deep fusion of a cross-platform soft decoding architecture, single-frame memory optimization, multi-thread parallel, GPU acceleration and other technologies, perfect balance of cross-platform, high performance and low memory is realized, and the development efficiency of the whole project is improved.

The invention relates to the field of artificial intelligence, and discloses a multi-modal memory system and method for intelligent interaction. The system comprises a multi-modal information acquisition module, a confidence evaluation module, a scenario aggregation module, a knowledge graph construction module, a hierarchical storage module, an intelligent retrieval module, an active verification module and a memory management module. The core problems that in the long-term user interaction process of an existing artificial intelligence system, multi-modal information management is fragmented, information credibility is not quantitatively evaluated, memory organization lacks semantic association, a retrieval mode is single and passive, and memory life cycle is not adaptively managed are solved. Finally, unified collection, quantitative confidence evaluation, scenario semantic organization, associative intelligent retrieval and adaptive memory optimization of multi-modal information are realized, high-quality and high-efficiency long-term memory support is provided for scenes such as intelligent assistants, smart home, medical health and educational training, and the user experience and practical value of an artificial intelligence system are remarkably improved.

The invention belongs to the technical field of artificial intelligence, and discloses a large model memory management and optimization system and method based on a vector data lake. Comprising a multi-dimensional memory vectorization coding and hierarchical management module, an intelligent retrieval and dynamic updating module, and a memory optimization strategy and interactive evaluation and cyclic adjustment optimization module based on a reflection and backtracking mechanism. According to the large model memory management and optimization system and method based on the vector data lake, a multi-dimensional and hierarchical memory storage and dynamic retrieval framework is constructed, collaborative management of long-term / short-term memory is supported, efficient reasoning, knowledge reuse and continuous learning ability of an intelligent agent in a complex dynamic task are achieved, and the system and the method have the advantages that the system and the method are simple and convenient to operate. Therefore, the robustness, the adaptability and the task migration capability of the intelligent system are remarkably improved.

The invention provides a memory management method, a memory controller and an effective data positioning circuit unit. The method comprises the following steps: determining a plurality of candidate virtual blocks according to respective valid data count values of a plurality of virtual blocks; obtaining a mapping table bitmap corresponding to each candidate virtual block, determining a plurality of target virtual blocks based on the mapping table bitmaps, and generating a corresponding integrated mapping table bitmap; obtaining one or more sub-mapping tables according to the integrated mapping table bitmap; and generating an effective data table according to the physical address in the sub-mapping table and the physical address contained in the target virtual block. According to the method, the candidate objects for garbage collection are subjected to secondary screening in a hardware acceleration mode to determine the optimal target combination, and the mapping relation of the candidate objects is integrated to obtain the physical address of the corresponding effective data, so that the effective data table is made to improve the efficiency of garbage collection operation, and the reading burden of a storage device is reduced.

The invention relates to the field of industrial Internet of Things, and discloses an industrial scene-oriented multi-protocol compatible IoT acquisition and intelligent operation and maintenance system, which comprises a configuration management module used for analyzing an encrypted configuration file and dynamically loading a communication protocol driving program by using a reflection mechanism; the data acquisition module is used for establishing connection with industrial field equipment by utilizing the loaded communication protocol driving program and acquiring original data; the data processing module is used for calculating a semantic entropy value of the original data in the time window and marking a data value grade according to the semantic entropy value; the data storage module is used for executing differentiated storage operation on the data according to the data value grade; a memory management module and a data forwarding module. By constructing a data acquisition module comprising a software and hardware dual-mode interface and combining a self-adaptive driving correction mechanism based on physical layer signal feature analysis, the effect of still maintaining communication connection when standard protocol driving fails or equipment aging parameters drift is achieved.

The invention is suitable for the technical field of terminals, and particularly relates to a memory management method, electronic equipment and a computer readable storage medium. According to the method, when it is detected that the first application is switched to the background state, the electronic device can determine that the first application does not need to display an image and does not need to carry out graphic rendering, that is, it can be determined that the first application does not need to use a GPU physical memory currently. At the moment, the electronic equipment can store the first content stored in the first GPU physical memory corresponding to the first application to a disk, and can release the first GPU physical memory, so that the occupation of the GPU physical memory is reduced, the remaining available physical memory of the electronic equipment can be increased, the response speed of the electronic equipment to user operation is increased, the fluency of the system is improved, and the user experience is improved. And user experience is improved.

The invention provides a multilevel modular personal memory large model and a construction method thereof. The model comprises a receiving module for receiving information input by a user, converting the information into a unified text and language vector, storing a personal historical language vector and context information, extracting key features, analyzing intention, retrieving by using Elasticsearch and returning related context information, and generating a reply according to the related context information. According to the method, the multi-modal data processing capability is integrated, the traditional limitation is broken through, and image, voice and text associated storage is supported; constructing a dynamic memory enhancement mechanism to prevent important information from being covered; then an intention analysis module is innovated to accurately capture deep demands; the correlation and the adaptability of memory fragments are improved by adopting a multi-modal retrieval and graph structure correlation technology; and finally, through a language habit adaptation and logical reasoning module, the output conforms to the style of the user, potential requirements are deduced, a whole-process closed loop is formed, and breakthrough is achieved in the aspects of memory management, intention understanding and interaction naturalness.

The invention discloses a graph-driven self-attention compressible memory management method, and belongs to the technical field of penetration testing. According to the graph-driven self-attention compressible memory management method, an attention graph is constructed, semantic dependence and attention flow directions among multiple Agent nodes are accurately described by utilizing edge weights, the relevance and the accessibility of information are improved, and in the aspects of screening and loading, the expandability of the memory is improved. According to the method, key memory fragments are screened on the basis of concerned graph edge weights and task related features, only the loaded information is loaded, redundancy is effectively eliminated, the data volume processed by a system is reduced, and the system operation efficiency is improved; the semantic compression module can perform abstract processing on historical information based on various elements, and on the premise of ensuring that key semantic information is not lost, the abstract length is controlled within the window limit, so that the model can normally process information, and the adaptability and stability of the system to different data volumes are improved.

The invention relates to the field of memory management, and provides a memory allocation method and device for a mobile terminal system. The method comprises the following steps: dividing a DDR (Double Data Rate) memory in the mobile terminal into a plurality of physical blocks through a memory management unit; uniquely numbering the plurality of physical blocks, and establishing a mapping relationship between the virtual address used by the mobile terminal system and the physical numbers corresponding to the plurality of physical blocks through the memory management unit; after receiving a memory allocation request, judging whether a target physical block indicated in the memory allocation request belongs to a single channel of the 16-bit memory; and if the target physical block does not belong to the single channel of the 16-bit memory, remapping a physical number in the memory allocation request into an equivalent physical number in a single channel range, and importing the to-be-allocated data into a corresponding 16-bit memory space through a memory management unit based on the mapping relationship and a virtual address corresponding to the equivalent physical number. The memory allocation efficiency is improved by realizing optimal allocation of the 16-bit memory.

Provided is an accelerator for accelerating batch large language model (LLM) inference via key-value cache compression. An accelerator, according to one embodiment, may comprise a plurality of compute cores. Here, each of the plurality of compute cores may comprise a plurality of processing units for processing an LLM inference operation on a per-token basis, and a direct memory access unit for managing operations of reading weights from a memory and writing key-value activation data back to the memory. In addition, the direct memory access unit may comprise a compression engine for managing online compression of the key-value activation data when the key-value activation data is written to the memory, a decompression engine for decompressing the compressed key-value activation data retrieved from the memory, and a memory management unit for managing reading and writing of the compressed key-value activation data in the memory.

The invention discloses an intelligent agent memory management method and system. According to the method, deep semantic organization of memory content is realized by defining a structured memory unit containing ternary metadata formed by a multi-label set, a multi-dimensional attribute set and a multi-level path; and receiving and processing a memory operation from any trigger event through a unified memory management framework. According to the method, the problems of low retrieval precision, poor data isolation between instances and tight coupling with service logic caused by flattening of a memory structure of an existing agent system are effectively solved, and the performance, the safety and the expandability of the agent system in enterprise-level application are remarkably improved.

The invention relates to a GPU-based adaptive compressed data direct query method, system and device and a medium, and the method comprises the steps: carrying out the data compression of input column data through employing an adaptive compression strategy, and obtaining a block-level structure suitable for GPU storage and calculation; loading the compressed data into a GPU memory, and performing Tile-level memory management by taking Tile as a basic scheduling unit; and according to the received query statement, executing direct query of the compressed data on the Tile level through the GPU, and outputting a query result. By designing a Tile-level direct query framework, a hardware-aware memory management and control flow coordination mechanism and a self-adaptive compression strategy, high-performance query execution is realized without decompression, and the computing potential of the GPU is fully released. The method can be widely applied to the technical field of big data processing.

The invention discloses a memory management method, which comprises a two-stage CXL memory sub-page sensing step based on pre-splitting: scanning a large transparent page based on a coarse-grained page table scanning mechanism, determining a large thermally transparent page, and putting the large thermally transparent page into a pre-splitting queue for management; pre-splitting a memory range related to the thermal transparent large page into a plurality of small pages; the page table items of the split small pages are utilized to achieve fine-grained scanning of the page table, and hot pages in the page table are obtained; migrating the hot page to a DDR memory node; and pushing the address area of the transparent large page out of a queue, and updating the page table item of each hot page.

The invention provides an address mapping method and device, electronic equipment and a chip, and relates to the technical field of memory management, the method comprises the following steps: obtaining a plurality of original memory addresses, at least two original memory addresses in the plurality of original memory addresses including the same memory bank address; based on an original address bit of each original memory address, determining a target address bit of each original memory address, the number of bits of the target address bits being a first preset number, and the first preset number being smaller than the number of bits of the original address bits; determining a mapping result of each original memory address based on the target address bit and a preset mapping relation; target memory addresses corresponding to the original memory addresses are determined based on the mapping result and the memory bank addresses in the original memory addresses, and the memory bank addresses included in the target memory addresses corresponding to different original memory addresses are different.

The invention discloses an intelligent dialogue memory management method and system based on the logistics field, and relates to the technical field of logistics intelligent dialogues, and the method comprises the steps: obtaining the historical logistics dialogue data of a target user, generating a user feature-interactive entity-semantic relationship initial knowledge graph based on a QWen2.5-32B large language model, and constructing a Milvus semantic vector library; monitoring a logistics dialogue flow in real time, and starting an adaptive memory management mechanism by taking three rounds of dialogue as a judgment threshold value; extracting user dialogue content attribute features and behavior preferences, and dynamically updating a user feature-interactive entity-semantic relationship initial knowledge graph and a Milvus semantic vector library; and triggering a logistics dialogue according to a target user, starting a multi-modal memory recall mechanism, cooperatively retrieving a user feature-interactive entity-semantic relationship knowledge graph and a Milvus semantic vector library, generating a personalized intelligent dialogue response, and realizing a logistics intelligent dialogue memory management closed loop. The beneficial effect of the invention is that the intelligent and personalized capabilities of the system are enhanced.

The application discloses a memory management method and device, a storage system and a computer readable medium, and relates to the technical field of mobile terminals. The method comprises the following steps: determining a target pre-erasing mode in an automatic mode and an active mode. In the automatic mode, the memory automatically performs a pre-erasing operation on the memory. In the active mode, the memory performs a pre-erasing operation based on an instruction sent by the host. A pre-erasing operation is performed on the memory based on the target pre-erasing mode. Therefore, for the automatic pre-erasing mode and the host-triggered pre-erasing mode, one of the two modes can be freely selected to perform a pre-erasing operation in the automatic mode and the active mode, so that the pre-erasing operation is more reasonable.

Methods, systems, and devices for adaptive memory status reporting are described. For example, the memory system may store one or more status indicators to a register for a host system to access. The status indicators may represent an estimated completion time for a quantity of access commands in a command queue of the memory system. The host system may use the status indicators to detect an unresponsive memory system. For example, the host system may query the register to determine whether to continue to wait for the commands to complete or to issue an abort command. Additionally, or alternatively, the host system may initiate memory management operations to assist the memory system in completing executing the commands.

The embodiment of the invention provides a memory management method, electronic equipment, a memory management device and a storage medium, and relates to the technical field of computers, the method comprises the following steps: obtaining a memory allocation request initiated by an application program; the migration type of target data aimed at the memory allocation request is determined as a target migration type, and the migration type comprises a migratable type and a non-migratable type; determining whether a free space capable of accommodating target data exists in the page block of which the migration type is the target migration type or not, wherein the migration type of the page block is the migration type of data capable of being stored in the page block; if yes, storing the target data into the page block; if the page blocks do not exist, searching the page blocks of which the migration types are idle types, storing the target data in the searched page blocks, updating the migration types of the page blocks storing the target data to the target migration types, and not storing the data in the page blocks of the idle types. By applying the scheme provided by the embodiment of the invention, the problem of memory fragmentation can be solved.

The invention relates to an Agent operation system and method oriented to a long-term task and driven by a task state, and belongs to the field of artificial intelligence. A state system is used for explicit modeling and unified management of an overall cognitive state; the context constructor is used for constructing a structured context matched with the current decision stage; the decision engine is used for outputting a structured decision intention; the execution and tool layer converts the structured decision intention output by the decision engine into executable operation; the memory management module is used for realizing memory enhancement, fading and hierarchical storage according to the access frequency of the context builder; and the verification and control module is used for performing consistency verification and process control on the decision result and the execution result. According to the method, structured management of task states, accurate control of task mainlines, controllability of context construction and interpretability and auditing of decision-making processes can be realized, and reliability and applicability of an Agent system in long-term, multi-tool and complex task scenes are improved.