53 results about "Context processing" patented technology

One aspect of the instant disclosure may provide a system and method for processing scatter-gather direct memory access (S-G DMA) instructions. During operation, the system may receive an S-G DMA instruction associated with a message and gather instruction context for the S-G DMA instruction. An S-G DMA processor may process the S-G DMA instruction based on the gathered instruction context and determine whether there exists a pending S-G DMA instruction associated with the message. In response to the presence of the pending S-G DMA instruction, the system stores the instruction context in a hot context cache at an address corresponding to the pending S-G DMA instruction. In response to the absence of the pending S-G DMA instruction, the system stores the instruction context in a cold context cache.

The invention provides a large model context processing method and device, computer equipment, a storage medium and a computer program product. The method comprises the steps of obtaining a new action step request, updating a context view tree and a source interaction data tree based on the new action step request, and determining a pressure coefficient according to the updated context view tree; according to the pressure coefficient, determining whether a to-be-compressed node exists in the updated context view tree or not; if the to-be-compressed node exists, combining the updated source interaction data tree and the updated context view tree to perform back-to-source compression on the to-be-compressed node so as to update the updated context view tree and the updated source interaction data tree, and returning to the step of determining the pressure coefficient according to the updated context view tree until the to-be-compressed node does not exist; and based on the updated context view tree, expanding the context view tree into context content and inputting the context content into the large language model. By adopting the method, the large language model can accurately understand and respond to the new action step request.

The invention provides a retrieval recall method and system fusing document fragmentation and authority control, and relates to the technical field of artificial intelligence and information retrieve.The method comprises the steps that a role of a user is obtained according to the identity of the user, and a corresponding authority strategy is determined in combination with a preset multi-level authority mapping relation; the permission strategy comprises a label allowed to be accessed, a data set and a document set; the query request is converted into a semantic vector, and initial candidate document fragments are retrieved from a vector database in combination with an associated data set or tag identifier; according to the method, the candidate results are subjected to access control filtering on the basis of the authority strategy, and the final retrieval recall result is generated, so that role-based fine-grained access control is realized while high retrieval precision is ensured, the problems of rough authority management and weak context processing capability in traditional retrieval are solved, and the security and efficiency of retrieval are remarkably improved.

The invention discloses a multi-round dialogue system and method based on conversion from a natural language to an SQL, and relates to the technical field of artificial intelligence, and the system comprises a user interaction module which is used for supporting a user to input a query problem and displaying a corresponding SQL query result; the context processing module is used for screening the effective dialogue history of the current question and re-integrating the effective dialogue history into a complete dialogue; the ambiguity processing module is used for identifying an entity from a dialogue history to define a user intention, processing a fuzzy keyword in a current question and guiding the user to complement a query condition; the re-splicing module is used for reconstructing a complete dialogue according to a time sequence, eliminating ambiguity, generating a summary through a large model, and combining the summary with a current problem to construct a complete problem; the SQL generation module is used for performing semantic analysis on the complete problem and outputting an SQL query instruction with the highest matching degree; and the execution and rendering module is used for executing the instruction on the target database and returning a query result. The conversation processing capability can be improved.

The present invention provides a system and method for multi AI agent driven data processing in enterprise application developed by codeless platform. The invention includes one or more AI agents configured for processing one or more input received on conversational assistant interface. The invention includes triggering contextual processing of received input by orchestration agent interacting with multi-AI agents for executing a task identified from the received input.

The invention relates to the technical field of artificial intelligence, in particular to an adaptive context processing method and system for a large language model. The method comprises the following steps: collecting multi-source candidate context segments related to current user query; performing utility evaluation on each candidate fragment, outputting a quantized utility score, and performing multi-granularity compression on each fragment based on the utility score and a preset quality constraint to obtain a compressed fragment set; taking the maximization of the total utility score as a target, taking the condition that the total token cost is less than or equal to the global token budget as a constraint, and carrying out optimization selection to obtain an optimal context combination; and inputting the optimal context combination into a large language model to generate an output result for the current query. By applying the method, the context window management of the large language model can be optimized, the adaptive context processing capability of the information utilization efficiency is improved, and the problems of context window limitation, value judgment and the like are solved.

The invention relates to the technical field of psychological intervention adjustment, in particular to a dynamic structured self-adaptive training method which is applied to a training framework comprising an acquisition terminal, a processing terminal and an execution terminal. The acquisition terminal is used for acquiring a physiological signal, a behavior signal and an environment context of an individual, the processing terminal fuses the information and calculates an anxiety index and an uncertainty probability value, and when the uncertainty probability value is lower than a preset threshold value, an initial task graph is constructed according to a target task and the environment context. The initial task graph is subjected to structure and parameter updating through a preset rule generator in combination with a reward function and an anxiety index, an individualized task graph is generated, cross-age migration is further carried out based on physiological and behavior signals, an execution graph bound with an execution terminal is obtained, and the execution terminal carries out intervention according to the execution graph. The problem that an existing training method is difficult to give consideration to behavior complexity and psychological bearing capacity is solved, and training individuation and stability are improved.

The invention relates to the technical field of nuclear power station monitoring, and relates to a method and system for rapidly identifying small foreign matters in a reactor pool, and the method comprises the steps: firstly, processing a reference image and a matched image based on an RAFT optical flow method combined with an attention mechanism to obtain an optical flow graph, and then carrying out the segmentation of the optical flow graph to extract a motion region; and the motion tends to be mapped to the matched image to obtain a foreign matter recognition result. According to the invention, by adding an attention mechanism, the context processing capability of the RAFT method is enhanced, and the optical flow of the water surface ripples can be calculated, so that the precision of optical flow estimation is improved.

The present disclosure relates to techniques for improving inference efficiency and memory utilization in transformer-based neural networks. A modified architectural design is introduced that decouples key and value matrix generation from inter-layer dependencies, enabling statically computed or parallelizable projections across layers. The disclosed approach may eliminate the need for layer-wise prefilling, support linear-time inference, and substantially reduce the memory footprint associated with key-value (KV) caching. The disclosed architecture may use shared or layer-specific projections, with a single KV-cache serving all or subsets of layers. In some embodiments, a non-linear transformation (e.g., implemented via a feed-forward network), may preprocess input embeddings prior to query, key, and value generation. A lookup table of transformed embeddings may be precomputed to further accelerate inference. The disclosed system can enhance scalability and may allow deployment of large models on resource-constrained hardware, offering practical benefits for latency-sensitive applications and long-context processing in transformer-based models.

A system includes an image signal processor (ISP). The ISP is a discrete hardware unit and includes a line buffer, a memory controller, and one or more image processors. The memory controller is configured to assign blocks within the line buffer to a plurality of contexts, and for each of the contexts: receive one or more lines of image data associated with the respective context, and storing the one or more lines of the image data associated with the context in a respective block assigned to the respective context. The one or more image processors of the ISP are configured to process, for each respective context of the plurality of contexts, the one or more lines of the image data stored in the line buffer associated with the respective context.

The application relates to the technical field of recruitment, and discloses a real-time position intelligent recommendation system based on an instant communication context, which comprises an instant communication access module, a real-time context processing engine, a multi-dimensional semantic understanding module, a dynamic position matching engine, a real-time recommendation pushing module, a data storage module, a permission management module, a business opportunity conversion module and a compliance processing module, and each module cooperatively works; the application solves the context split problem, realizes seamless experience of "communication is data", through efficient communication architecture of the instant communication access module and the back end, real-time capture of bidirectional chat messages of HR and candidates, customers, breaking of the split state of the communication scene and the recommendation system in the prior art, inclusion of each interaction into the recommendation consideration, complete capture of dynamic demands and implicit demands in the communication process, and avoidance of demand information omission.

The application discloses a long context processing robust enhancement method and system based on neuron dynamic adjustment, and belongs to the technical field of text sequence processing, and comprises the following steps: embedding a nerve regulation unit in each Transform layer of a large language model used for a long context processing task; and the nerve regulation unit sequentially performs the following steps: dynamically activating abnormal perception, that is, adaptively calculating an abnormal detection threshold of a current layer according to activation value statistical characteristics of the current layer; activating a hard-gate truncation, that is, generating a binary hard mask according to the abnormal detection threshold, and the binary hard mask realizes deterministic isolation of abnormal components in the activation value while keeping normal components unchanged; and performing hierarchical response compression, that is, performing learnable dimension-level linear compression on the isolated abnormal components, re-integrating the compressed abnormal components and the normal components through gate fusion, and outputting an activation tensor after suppressing the abnormality. The application can prevent rank collapse and gradient oscillation, and enhance the robustness of long context task execution.

This application relates to the field of natural language processing, and more particularly to a method for ultra-long context tensor parallel incremental attention and differential compression transmission. It employs a multi-level precision sliding window attention mechanism, dynamically allocating precision levels such as fp16 / int8 / int4 based on the distance between the window and the current computation position, and the information entropy of the window content. A tensor parallel incremental attention engine enables cross-GPU / node incremental attention computation and asynchronous All-Reduce synchronization, improving computational efficiency. Semantic differential compression transmission technology is used, transmitting only the differential data based on similarity detection, significantly reducing cross-node transmission overhead. Combined with an extreme performance optimization and security control strategy, a balance between performance and security is achieved through hardware monitoring and dynamic power consumption / frequency adjustment. This invention can significantly reduce memory usage and transmission overhead in ultra-long context processing, improve computational efficiency and hardware utilization, enhance the understanding ability of large models for long texts, adapt to heterogeneous hardware environments, and is suitable for large model inference and training scenarios.

Surgical planning based on a three-dimensional (3D) model includes obtaining a 3D model of an anatomical region, wherein the anatomical region includes the patient's anatomical features; identifying at least one deformity, wherein the at least one deformity is identified with respect to a target anatomical value of the patient's anatomical features, wherein the target anatomical value includes a desired range into which the anatomical measurement should fall; and using the anatomical measurement, determining at least one planar correction to be made to at least one anatomical structure of the patient. Generating a 3D model using two-dimensional (2D) imaging data may include performing anatomical context processing, which includes annotating the model, contours, edges, or surfaces of the anatomical structure, as well as performing 2D-to-3D reconstruction to optimize the orientation, placement, and scale of the 3D digital volume modeling the anatomical structure, resulting in a 3D anatomical representation of the anatomical region as a 3D model.

A dialog understanding device includes a sound collection module configured to collect a sound signal, a contextual processing module, and a dialog system configured to perform a dialog with a human. The contextual processing module includes a plurality of layers for processing information obtained from the sound collection module. A fallback processing module of a case where a predetermined process has not succeeded for the collected sound signal is provided on each layer of the plurality of layers. A process of the next layer is performed after a fallback corresponding to a fallback process performed by the fallback processing module is performed. A sound signal obtained when the contextual processing module completes a process is input to the dialog system.

The present disclosure provides techniques for seamless roaming with uplink / downlink context transfer. A serving AP transmits a sequence of downlink data units for a TID to a client device, the sequence of downlink data units having sequence numbers falling within a transmit window for the TID. The serving AP receives, from the client device, a roaming request identifying a target AP. In response, the serving AP sends a roaming context message to the target AP, comprising at least one of a SSN of the transmit window corresponding to the TID, or a NSN for the TID, where the NSN is a first sequence number to be assigned for downlink data units of the TID transmitted by the target AP to the client device.

The invention discloses a KV cache intelligent scheduling and semantic structure compression method and system based on a knowledge graph, and the method comprises the steps: firstly, extracting entities and relationships from an input text to construct the knowledge graph, and then employing a graph embedding and structure compression algorithm to aggregate nodes with similar semantics into a high-density semantic KV item, so as to eliminate redundancy, and improve the KV cache intelligent scheduling and semantic structure compression efficiency. And then constructing a semantic scheduling graph, and dynamically loading the most relevant semantic sub-graph cache through a semantic migration mechanism. According to the method, the problems of semantic redundancy, lack of structural understanding, poor dynamic adaptability and the like in a traditional KV cache are effectively solved, the cache utilization rate, semantic consistency and long-range context processing capacity are remarkably improved, and the method is suitable for scenes such as large language model reasoning acceleration and intelligent dialogue.

The invention provides a two-level context caching and eviction for distributed aggregation DMA (direct memory access). One aspect of the present disclosure may provide a system and method for processing a decentralized aggregated direct memory access (S-G DMA) instruction. During operation, the system may receive an S-G DMA instruction associated with a message and aggregate an instruction context of the S-G DMA instruction. The S-G DMA processor may process the S-G DMA instruction based on the aggregated instruction context and determine whether there is a pending S-G DMA instruction associated with the message. In response to the presence of the pending S-G DMA instruction, the system stores an instruction context at an address in the thermal context cache corresponding to the pending S-G DMA instruction. In response to the absence of the pending S-G DMA instruction, the system stores an instruction context in a cold context cache.

The present application relates to a kind of IPTV terminal edge computing method and system, it is related to IPTV terminal technical field, including the following steps, IPTV terminal edge computing method first analyzes user interaction behavior, obtains behavior mode.Based on this mode, system carries out dynamic matching and context processing to content request, generates content preference matrix.Then, the task scheduling of local edge node is optimized using the matrix.System subsequently carries out hardware resource mapping, obtains equipment adaptation parameter, and accordingly pipeline adaptation is carried out to video decoder, generates local rendering instruction.Finally, based on these instructions, video frame buffer is rendered asynchronously, and the output of smooth video play stream is solved.The problem that traditional centralization IPTV system often exists network delay, service quality decline when facing large-scale user access.

The computing device includes a processor and a context switcher. The context switcher includes a data structure module for storing a context, which includes the starting address of a thread and values ​​associated with that thread. In an idle state, the processor loads a first context for a first thread from the corresponding data structure module in synchronization with a first cycle of a clock signal. In an active state, the processor processes the first thread according to the first context. Furthermore, in an active state, the processor provides a second context for a second thread to the context switcher. This second thread follows and will be processed in parallel with the first thread. The second context is stored in a subsequent data structure module within the context switcher's data structure module, which corresponds to a subsequent processor for processing the second thread.

A system and method are provided for encoding and decoding video for machine consumption, wherein bandwidth is reduced by filtering feature layers determined to be redundant or less relevant at an encoder site. The video encoder includes a neural network front end that receives image data and generates multiple feature layers. A layer context processor and a redundancy layer identifyer work together to evaluate the context of objects in the image data and determine the relevance of multiple feature layers at the decoder site to the machine task. The layer filter then performs at least one of the following to generate a filtered set of layers: removing redundant layers and scaling layers identified as less relevant to the machine task.

The application discloses a document query method and device, equipment, storage medium and computer program product, relates to the technical field of natural language processing, and the method comprises the following steps: generating a target user query based on a current user query input by a user and a historical question and answer pair sequence of the user; determining a target compressed text paragraph in a preset document semantic tree according to a target paragraph index matched by the target user query, the preset document semantic tree storing a mapping relationship between compressed text paragraphs in a target document and paragraph indexes; and generating a document query result corresponding to the current user query based on the target compressed text paragraph. According to the application, the compressed text paragraph can be matched in the document semantic tree according to the user query, and the document query result is generated based on the compressed text paragraph, so that the technical problem that the accuracy of the query result is not high due to the fact that the retrieved document paragraph exceeds the context processing capability of the model in the existing document query method is solved.

The invention provides a dialogue processing method and device, electronic equipment and a computer readable storage medium. The dialogue processing method and device are applied to a man-machine dialogue scene. The method comprises the steps of obtaining input information of a current dialogue and dialogue information of a historical dialogue; classifying the input information to obtain a classification label of the input information; performing context processing on the input information, the classification tag and the dialogue information to obtain a first prompt word of the current dialogue; and based on the first prompt word, performing dialogue prediction processing on the input information to obtain first reply information of the current dialogue.

Embodiments provide a computer that includes one or more processors and instructions stored on one or more memory devices. The one or more processors executes the instructions to monitor, using one or more extractor application programming interfaces (APIs), a plurality of streams associated with user interactions with the computer, each of the plurality of streams including one or more events or activities associated with the user interactions, the user interactions being interactions with one or more different applications or platforms of the computer. The processor may extract contents and contexts from the plurality of streams using the one or more extractor APIs. The processor further determines one or more workflows associated with the user, the one or more workflows including the extracted contents and contexts, the workflows indicating the user interactions with the computer during a time series comprising a plurality of timestamps associated with times at which the user performed the interactions with the computer. The processor stores the one or more the workflows at the user computer. The processor determines one or more chunks of the one or more workflows, each chunk being a subpart of one of the one or more workflows and corresponding to a set of the plurality of timestamps associated with the subpart, each chunk including the content and contexts of the one or more workflows associated with the set of the plurality of timestamps. The processors evaluates the one or more chunks, the evaluating creating one or more inferred actions using the content and contexts of the one or more workflows associated with the set of the plurality of timestamps and automates one or more subsequent workflows associated with the user based on the one or more inferred actions.

The present disclosure provides techniques for seamless roaming with uplink / downlink context transfer. A first AP MLD of a SMD establishes an association between the SMD and a non-AP MLD, receives UL MPDUs from the non-AP MLD, each UL data MPDU having a corresponding SN, transmits content of the UL data MPDUs received from the non-AP MLD to a distribution system, maintains UL context information of the association, receives a roaming request from the non-AP MLD, transfers UL context information to the second AP MLD, the UL context information comprising a SN of the last UL data MPDU received from, and acknowledged to, the non-AP MLD and forwarded to the distribution system, and transmits a roaming response to the non-AP MLD, the roaming response comprising an indication of the SN of the last UL data MPDU received from, and acknowledged to, the non-AP MLD and forwarded to the distribution system.

The invention relates to a data generation method and a related product. The data generation method comprises the steps of obtaining a target text, wherein the character number of the target text is greater than a preset value; the target text and a first prompt word are input into a first language model, a target question related to the target text is obtained, and the first prompt word is used for guiding the first language model to output the question related to the target text based on the target text; inputting a second prompt into the first language model to obtain a target answer for answering the target question, the second prompt being used for guiding the first language model to output the answer of the target question based on the target text; and obtaining target data based on the target text, the target question and the target answer, the target data being used for training a language model. By adopting the method, the target data for training the language model can be generated, so that the language model obtained based on target data training has the capabilities of understanding a long-length context and processing long-sequence information.

A method performed by a first access point (AP) in in a wireless local area network (WLAN) is provided. The method comprises receiving, from a station (STA), a roaming request associated with a roaming procedure for roaming from the first AP to a second AP, wherein during the roaming procedure, the first AP transmits data for the STA and a distribution system (DS) remapping procedure is initiated; determining a value of a latest SN of a plurality of SNs associated with traffic identifier (TID) information of the data of the STA at the first AP when the roaming procedure is initiated; and transmitting, to the second AP, information associated with a value of a first SN associated with the TID information at the second AP based on the determined value of the latest SN at the first AP when the roaming procedure is initiated.

Some embodiments are directed to systems and methods that generate and control workflows. In one aspect, a computer system includes one or more processors and memory. The computer system detects one or more user actions requesting context data associated with a workflow, retrieves the context data from the memory, and receives a user response associated with the context data. The computer system applies a context processing model to process the context data and generate model output data. The computer system generates a workflow controlling instruction based on the user response and the model output data. The computer system at least partially controls the workflow using the workflow controlling instruction.

Embodiments of the present application provide a context processing method and device, electronic equipment and storage medium, relating to the technical field of large language models. The method guarantees the timeliness of the response through a threshold triggering mechanism during the operation of an industrial large language model inference engine, realizes "zero loss of key information and high compression rate of non-key information" with the help of structured perception and dynamic importance evaluation, and achieves high compression rate and inference acceleration through a hierarchical compression strategy. Finally, through recoverable reconstruction, the compressed context is strictly compatible with the input specification of the large language model, and fully supports breakpoint continuation, full audit and lossless backtracking of original data, thereby solving the multiple contradictions of context management in accuracy, real-time performance, compliance and edge adaptability in industrial scenarios.

A system comprises an image signal processor (ISP). The ISP is a discrete hardware unit and includes a line buffer, a memory controller, and one or more image processors. The memory controller is configured to allocate blocks within the line buffer to a plurality of contexts, and for each of the contexts: receive one or more lines of image data associated with a respective context and store the one or more lines of the image data associated with the context in a respective block allocated to the respective context. The one or more image processors of the ISP are configured to, for each respective context of the plurality of contexts, process the one or more lines of the image data associated with the respective context that are stored in the line buffer.