401 results about "Bandwidth utilization" patented technology

The invention relates to the technical field of communication networks, in particular to a dynamic resource allocation method and system for an industrial 5G network. The method comprises the following steps: acquiring industrial terminal data; identifying network congestion nodes based on industrial terminal data; detecting a link congestion fault according to the network congestion node, and determining a bandwidth utilization rate; allocating spectrum resources based on the bandwidth utilization rate; performing frequency spectrum slicing on the frequency spectrum resources, and determining a flow load amount; adjusting load balance according to the traffic load, and generating a load resource report; dynamically adjusting time slot allocation by using a load resource report, and determining a task priority; task scheduling simulation is executed according to the task priority, and if congestion occurs, a congestion snapshot is stored; reducing scheduling conflicts according to the congestion snapshots, releasing bandwidth resources, and determining available network slice capacity; power resources are dynamically adjusted based on available network slice capacity. The bandwidth utilization rate and the task scheduling efficiency of the industrial 5G network are improved based on the communication network technology.

According to the automatic driving data transmission scheduling method and device, the equipment and the readable storage medium, dynamic priority weights are generated in a mode of combining reinforcement learning and a rule engine, and bandwidth demand changes, historical transmission performance and queue occupation states of all sensors can be adapted in real time; the problem that a low-priority sensor cannot acquire the bandwidth at a high load is effectively avoided, and meanwhile, efficient utilization of bandwidth resources can be realized through dynamic weight adjustment when a high-priority sensor is idle, so that the bandwidth utilization rate in a vehicle-mounted multi-sensor data transmission process is remarkably improved; the real-time performance and reliability of data transmission required by automatic driving decision making are guaranteed, and then the overall performance of an automatic driving system is improved.

The invention discloses a bandwidth dynamic allocation method and device, a program product and a storage medium, and relates to the technical field of computers. According to the method, bandwidth use data of a network outlet are collected in real time, the residual bandwidth capacity of the outlet is determined, and the distributable dynamic bandwidth capacity of each interface is calculated by combining an interface weight parameter and an interface stability coefficient calculated based on bandwidth use fluctuation characteristics, so that the residual bandwidth of the network outlet can be reasonably distributed. Meanwhile, a dynamic additional bandwidth mechanism is introduced outside the fixed basic bandwidth, and the dynamic bandwidth allocation weight is determined based on the bandwidth use stability of the authentication terminal and the current bandwidth use state, so that the terminal is not limited by a preset fixed bandwidth upper limit any more, more bandwidth resources can be obtained when the exit bandwidth is sufficient, and the bandwidth utilization rate of the terminal is improved. Therefore, the utilization efficiency of bandwidth resources is improved.

The invention discloses a communication-detection integrated sonar signal design method and device based on multi-pulse composite modulation, and the method comprises the steps: dividing a total frame period into a detection sub-frame and a communication sub-frame, enabling the detection sub-frame to achieve the high-resolution distance measurement through employing a bidirectional linear frequency modulation (BILFM) signal, a communication sub-frame adopts a multi-frequency multi-phase composite modulation (MTPM) signal to realize high-bit-rate data transmission; after band filtering, orthogonal demodulation, low-pass filtering and down-sampling preprocessing are carried out on detection sub-frame echoes, distance and speed information is extracted by adopting fast-time low-sampling two-dimensional fast Fourier transform (2D-FFT), unambiguous Doppler frequency is recovered by utilizing the Chinese remainder theorem (CRT), and accurate speed measurement of long-distance and high-speed targets is realized; a communication subframe receiving signal is subjected to IQ demodulation after down-conversion, Costas loop carrier synchronization and matched filtering, and multi-frequency multi-phase coded data are extracted. According to the method, the time-frequency collaborative design of detection and communication is realized, the distance resolution and the speed measurement range are considered, and the signal bandwidth utilization rate and the underwater information transmission reliability are remarkably improved while the system complexity is kept low. The device comprises a processor and a memory, and an executable program for implementing the method is stored in the memory. The technology can be widely applied to the fields of unmanned underwater vehicles, ocean monitoring, underwater sensing networks and the like.

The invention relates to an activation value re-segmentation communication method for a heterogeneous AI acceleration chip server. The activation value re-segmentation communication method comprises the following steps: scanning through chips, network interface cards and PCIe levels in the AI acceleration chip server, extracting network topology information of the server, calculating delay and bandwidth from each chip to each network interface card, and generating an affinity matrix; based on the affinity matrix, each chip is bound to a network interface card in an exclusive or shared mode through maximum matching; based on the server network topology information, the chip-network interface card affinity and the network interface card and PCIe single line bandwidth, activation value communication mode matching is carried out, and communication modes comprise a transmitting / receiving mode, a transmitting / receiving + collecting mode and a transmitting / receiving + broadcasting mode; and double-level multi-channel communication between the chip and the network interface card is realized. The method has the advantages of remarkably improving the bandwidth utilization rate, being highly adaptive to topology difference, being high in heterogeneous compatibility and the like.

The invention discloses a model performance test method and device, electronic equipment and a storage medium, and relates to the technical field of artificial intelligence. The theoretical maximum lexical throughput of a target large language model is calculated based on the video memory bandwidth of a graphics processor, the model parameter quantity, the byte number corresponding to the quantization precision and the video memory bandwidth utilization rate; meanwhile, the benchmark performance throughput is obtained, a theoretical corresponding first concurrency number is calculated in combination with the theoretical maximum lexical unit throughput and the concurrency competition loss coefficient, then the model test is executed based on the first concurrency number to obtain the actual maximum lexical unit throughput and a corresponding second concurrency number, and a model performance test result is generated. The problems that in the prior art, due to the fact that manual testing is conducted depending on manual intervention, a continuous approaching attempt mode is adopted, a reasonable test starting point is not deduced in combination with hardware core bottlenecks and key parameters, evaluation is time-consuming and labor-consuming, the result is prone to being affected by artificial factors, and accuracy and consistency are poor can be solved.

The invention discloses a multi-dimensional layered current limiting method based on spatio-temporal feature fusion. The method comprises the following steps: acquiring a hardware limit speed, and generating a global quota based on a reinforcement learning model; calculating a variable coefficient, and if the variable coefficient exceeds a preset critical value, introducing a conservative coefficient and generating a weight factor through nonlinear function mapping to suppress the micro network jitter; constructing a priority factor matrix based on the service priority and the service type, and obtaining a service priority factor; correcting the average bandwidth of the node by integrating the multi-dimensional dynamic weight factor and the service priority factor to obtain a suggested speed, inputting a historical performance index into the LSTM model to predict the load of the next period, and if the historical performance index exceeds a critical value, triggering connection migration before executing current limiting to realize active avoidance; otherwise, comparing and selecting the minimum value of the global quota, the suggested speed and the hardware limit speed as the final current-limiting speed. The problems of current limiting strategy lag and wide oscillation can be solved, and the bandwidth utilization rate and the system stability are improved.

The invention discloses a cross-chip communication method, a direct connection module and a system, which consumes virtual channel credit during arbitration and packet reorganization, supplements the virtual channel credit when unpacked and uploaded to a network-on-chip, ensures that packets are correctly interacted in cross-chip communication through the channel credit, and reorganizes the packets according to a CXS protocol in an arbitration mode. According to the invention, ordered transmission of a plurality of packets in one physical channel is realized, 'starvation 'can be effectively prevented, the packet of a certain virtual channel is effectively prevented from being indefinitely delayed, the bandwidth utilization rate is improved, and the delay and congestion risks of cross-core-particle communication can be reduced.

The invention provides a memory access optimization method and device for an aggregate communication operator and computing equipment, and relates to the technical field of graphics processor computing. The method comprises the following steps: identifying a memory access instruction sequence in a to-be-optimized set communication operator, wherein the memory access instruction sequence comprises a standard loading instruction and a standard storage instruction; replacing the standard loading instruction with a batch loading instruction, and replacing the standard storage instruction with a batch storage instruction; the batch loading instruction and the batch storage instruction are configured to execute batch operation on the continuous memory address space at the thread beam level and finally execute the batch loading instruction and the batch storage instruction, so that the emission number of memory access instructions can be reduced, the scheduling burden of a thread beam scheduler is relieved, and the scheduling efficiency is improved. And the execution efficiency and the bandwidth utilization rate of the ensemble communication operator in the distributed training of the graphics processor are effectively improved.

The invention discloses an atypical sampling bit width data transmission method and device, equipment and a medium. Comprising the steps of obtaining a target transmission protocol and constructing instances according to target transmission protocol requirements; calculating the number of temporary frame bytes according to the instance, and when the number of temporary frame bytes is not an integer, determining that the current scene is atypical sampling bit width transmission; and determining a transmission frame number based on the temporary frame byte number, generating aggregation reconstruction data according to the transmission frame number, and performing data transmission based on the aggregation reconstruction data. By constructing an instance, protocol rules and parameters which need to be followed by data transmission can be determined. An atypical sampling bit width transmission scene is determined by calculating the number of temporary frame bytes, and a scene basis is provided for subsequently adopting an aggregation reconstruction scheme in a targeted manner. By determining the number of transmission frames, the atypical sampling bit width data can meet protocol requirements after being aggregated and reconstructed, redundant bits do not need to be supplemented or the sampling rate does not need to be reduced, protocol compatible transmission of the atypical data is directly achieved, and meanwhile the effective bandwidth utilization rate and the data transmission efficiency are guaranteed.

The invention discloses a switch, system and method for data transmission based on a shared pool and a fan-out structure.The switch comprises a plurality of ports, the shared pool and a fan-out cache unit, and the ports comprise the first port and the second port; the first port receives data outside the switch, transmits the data to the shared pool and transmits a first data transmission request to the second port, and the first data transmission request comprises a data length; the second port receives the first data transmission request, and the target port generates a second data transmission request according to the first data transmission request, the data length and the remaining storage space of the target port so as to transmit the second data transmission request to the fan-out cache unit; and the fan-out cache unit schedules the data at the corresponding position in the shared pool according to the second data transmission request and outputs the data to the target port when the residual storage space of the target port is greater than the storage space required by the data length. The bandwidth utilization rate and the response speed of the data exchange network can be effectively improved, and congestion and delay are reduced.

The invention discloses a video data processing method, electronic equipment and a readable storage medium, relates to the technical field of video data transmission, and is applied to a video data processing system which at least comprises an intelligent sensing module and a dynamic transmission module. The video data processing method comprises the following steps: extracting metadata in video data through an intelligent sensing module; collecting network state data, and sending the network state data and the metadata to a dynamic transmission module; target data transmission parameters are determined through the dynamic transmission module based on the network state data and the metadata, and the target data transmission parameters at least comprise one of redundancy, frame priority, code rate, retransmission parameters and quantization parameters; and transmitting the video data based on the target data transmission parameter. According to the technical scheme, the bandwidth utilization rate, the transmission real-time performance and the anti-frame-loss effect in the video data processing process are improved.

The invention relates to the technical field of flow regulation and control, in particular to a docking station data flow intelligent regulation and control method and system.The method comprises the following steps of obtaining unit time throughput and timestamps of all interfaces, judging slope inversion and abrupt change to generate a task switching boundary, constructing a difference value sequence to judge rate stability change to generate a strategy switching identifier; and calculating a regulation coefficient correction bandwidth proportion in combination with the service level and the request frequency, distinguishing a task type adjustment period and a rate level, and constructing a scheduling structure body to generate a regulation instruction set. According to the method, the bandwidth utilization rate change sequence of the port operation data is constructed, slope inversion and change amplitude judgment are combined, accurate identification of task change nodes is achieved, bandwidth regulation and control are carried out in combination with the service level label and the port request frequency, bandwidth resource allocation better fits the task importance and the actual request frequency, and the bandwidth resource allocation efficiency is improved. Meanwhile, the window length and the rate gear are dynamically configured according to task interactivity and periodic stability, and bandwidth waste and unnecessary transmission delay are restrained.

The invention discloses a channel scheduling and bandwidth allocation method, system and device for cloud desktop communication, and the method comprises the steps: obtaining a data frame set, allocating a corresponding priority identifier according to a channel type to which each data frame belongs, and embedding the priority identifier of each data frame into a special field of a frame structure; the data frames are inserted into the corresponding sending queues according to the priority identifiers, each priority corresponds to one sending queue, the data frames are extracted from all the sending queues in sequence according to a preset scheduling strategy to be sent, and the preset scheduling strategy at least supports a priority preemptive mode or a weighted fair queuing mode; and collecting network state parameters in real time, and dynamically adjusting the bandwidth quota of the channel corresponding to each data frame through a feedback closed-loop mechanism based on the network state parameters. According to the invention, the rapid transmission of high-priority data frames is ensured, and the delay of real-time interaction is reduced; resources are flexibly allocated according to the network state, and the bandwidth utilization rate is improved.

The invention discloses a bandwidth allocation method and device for a namespace, and the method comprises the steps: receiving an operation request, and determining a target namespace for which the operation request aims in an SSD (Solid State Disk); obtaining the capacity of the target namespace and the total bandwidth of the NAND Flash end of the SSD, determining active namespaces when the operation request is received, and obtaining the sum of the capacities of all the active namespaces; wherein the active namespace refers to a namespace with data operation in a preset time period; obtaining a capacity ratio between the capacity of the target namespace and the determined sum of the capacities of all the active namespaces; and allocating bandwidth for the target command space according to the total bandwidth of the NAND Flash end and the obtained capacity ratio, and executing the received operation request according to the allocated bandwidth. Therefore, the overall bandwidth utilization rate is improved.

The invention provides a data compression method, apparatus and device, and a storage medium. The method comprises the steps of obtaining an original data stream transmitted through a serial port; determining the data type of the original data flow according to the original data flow; determining a corresponding target compression algorithm according to the data type; compressing the original data stream according to the target compression algorithm to obtain compressed data; the defects of low efficiency and even data expansion during data abrupt change in a fixed compression mode are effectively overcome, and the data transmission efficiency and the bandwidth utilization rate of serial port communication in a resource limited environment are remarkably improved.

The invention discloses a data traffic scheduling method and a communication system based on traffic characteristics and path selection, and the method comprises the steps: presetting a large flow path and a small flow path according to the historical state information of each path, and enabling the path blocking risk and the path stability of the large flow path to meet a preset first requirement; the path blockage risk and the time delay of the small flow path meet preset second requirements, and the path blockage risk, the path stability and the time delay are determined according to historical state information; predicting whether the data stream sent by the sending end is a large stream or a small stream according to the traffic characteristics of the data stream historically sent by the sending end, and if the data stream is the large stream, determining that the sending end is a large stream sending end and distributing the data stream to a large stream path; and if the small stream is predicted, determining that the sending end is a small stream sending end and distributing the data stream sent by the small stream sending end to a small stream path. The bandwidth utilization rate and the response speed of a communication system can be effectively improved, and congestion and delay are reduced.

The invention discloses a double-flow image transmission system and method based on a high-speed SerDes interface, and belongs to the technical field of high-speed image transmission, the system comprises a sending end and a receiving end, an image source A and an image source B of the sending end do not need DDR cache synchronization, and image frames are segmented, interleaved and packaged through a dynamic time slice distribution mechanism and then sent to a single high-speed SerDes channel; and a receiving end collects data and then de-interleaves and recombines the recovered image. The data encapsulation comprises a frame synchronization head, a block synchronization head and the like for identification, the dynamic time slice distribution mechanism alternately or independently sends image blocks according to the image preparation condition, and the image adopts a double-cache mechanism to avoid data coverage. The method improves the channel bandwidth utilization rate, saves hardware resources, is good in compatibility, supports various data stream combinations, reduces frame delay, and is suitable for single-channel multi-stream and multi-channel multi-stream transmission.

The invention provides a large model reasoning method and device, a medium and electronic equipment. In the method, a central processing unit can divide a batch processing data set of a to-be-executed reasoning task into sub-batch processing data sets and distribute the sub-batch processing data sets to a plurality of graphics processors, so that the graphics processors execute reasoning calculation based on the distributed sub-batch processing data sets respectively, and an execution result of the to-be-executed reasoning task is obtained. In the process, each graphic processor can preload the weight parameter of the next reasoning calculation layer from the storage space corresponding to the central processor and store the weight parameter into a video memory in the process of executing the reasoning calculation task corresponding to any reasoning calculation layer; therefore, when the calculation task of the current reasoning calculation layer is not completed, the graphics processing unit can load the weight parameters required by the subsequent reasoning calculation layer into the video memory in advance, so that storage is replaced by passage, and the bandwidth utilization rate of the video memory and the overall reasoning throughput are improved.

The invention provides a data forwarding method based on a vehicle-mounted passive optical network and a related device, and relates to the technical field of vehicle communication. In the data forwarding method, the first ONU gateway aggregates and packages the received multiple frames of messages into the same uplink GEM frame, so that the multiple frames of messages share the same GEM frame header, the frame header overhead in the low-speed data transmission process is effectively reduced, and the bandwidth utilization rate of the vehicle-mounted PON network is improved. Meanwhile, message grouping, uplink GEM frame packaging and uplink burst frame coding are directly completed through the first ONU gateway, the uplink burst frame is sent to the OLT, time delay and system resource consumption are reduced, the high real-time requirement of low-speed data (such as a sensor and a control signaling) in a vehicle-mounted scene is met, and therefore efficient and low-delay bearing of the low-speed data on a vehicle-mounted PON network is achieved.

The invention relates to the technical field of data transmission, in particular to a distributed transmission method and system for large model reasoning and training, and the method comprises the steps: carrying out the classification and priority marking of training data and reasoning request data; dynamically determining a data fragmentation strategy based on the classification result, the node resources and the network state, and generating metadata; a differential transmission strategy is adopted to transmit data, incremental transmission is adopted to update model parameters, and collaborative scheduling and monitoring are carried out through a scheduling center; carrying out load balancing and pipeline transmission on the reasoning request; dynamically optimizing a transmission strategy and task allocation based on continuously acquired performance data; the system comprises a dispatching center, a computing node and a network state sensing module, through classification, dynamic fragmentation and differential transmission strategies, the network bandwidth utilization rate is effectively optimized, and the transmission delay and communication overhead of distributed large model training and reasoning are reduced.

The invention discloses a data center load balancing system and method based on an SDN (Software Defined Network). The system comprises a switch monitoring module, a load balancing module and a flow table issuing module. The method comprises the following steps: periodically obtaining network global state information through an in-band network telemetry INT technology; calculating a global average priority based on the INT data to distinguish high and low priority data streams; dynamically calculating an optimal path in K shortest paths for avoiding congestion nodes for the high-priority flow, and selecting a path for the low-priority flow by adopting an improved ECMP algorithm based on link utilization rate weighting; and finally compiling the path strategy into a flow table item and issuing the flow table item to the switch. Through a priority-aware dynamic load balancing mechanism, the Hash collision and static defects of the traditional ECMP are overcome, the transmission delay of the high-priority data flow is effectively reduced, the bandwidth utilization rate of the low-priority data flow and the whole network flow transmission quality are improved, and the method is suitable for a large-scale data center network.

The invention discloses an interactive display method, system and device for a video and a frame extraction picture based on a configuration file, a medium and a program, and belongs to the technical field of video monitoring. The method comprises the steps of forming an initial configuration file based on power plant data, and obtaining monitoring point information according to the initial configuration file; the monitoring point information is matched with the target monitoring point information, and after matching succeeds, the priority of the target monitoring point is obtained; according to the priority of the target monitoring point, broadband resource allocation is carried out, and a resource allocation strategy is obtained; and performing video or frame extraction picture display of the target monitoring point according to the resource allocation strategy. According to the method, through dynamic resource scheduling, intelligent frame extraction adaptation and low-delay interaction, the bandwidth utilization rate is improved, the abnormal response is accelerated, the operation efficiency is optimized, and the multi-scene compatibility is enhanced.

According to the near memory DMA asynchronous controller, the data transmission method of the AI computing system, the medium and the product, a buffer does not need to be configured, the area of the near memory DMA asynchronous controller is reduced, the data carrying path is shortened, and the power consumption overhead in the data carrying process is reduced. Bus bit width address alignment is carried out in the near memory DMA asynchronous controller and then transmission is carried out, so that the utilization rate of the bus bandwidth is improved, and the current requirements of the AI chip are adapted with low logic overhead, low power consumption and high transmission efficiency; the asynchronous control mode can enable the calculation process and the data transmission process of the AI chip in the operation of the network model to be higher in parallel degree and higher in coupling degree.

The invention relates to the technical field of communication, discloses a mining intelligent voice and video interactive communication method and system, and aims to solve the problems of poor audio and video signal quality, low bandwidth utilization rate, lack of semantic understanding and interaction lag in a complex mine environment. The method comprises the following steps: acquiring an original signal through an intrinsic safety type audio and video terminal; voice enhancement is carried out by adopting sound source orientation estimation and a deep complex network; reconstructing a high dynamic range video based on the physical imaging model; realizing audio and video semantic alignment and key event extraction by using a lightweight spatial-temporal feature fusion network; the bandwidth is dynamically allocated by combining the event confidence coefficient and the channel state, and high / common priority data is transmitted in a graded manner; and ground and underground real-time communication is realized through a bidirectional interaction channel. According to the invention, the signal availability, the bandwidth efficiency and the emergency response capability are remarkably improved, and high-stability, low-delay and high-fidelity communication is guaranteed.

The invention discloses an industrial data transmission optimization management system based on service link quality, and relates to the technical field of data transmission. Weak delay oscillation and a periodic packet loss rule are reserved through multi-scale processing, so that the system can discover precursor anomalies which are difficult to capture by a traditional monitoring mechanism; minute-level even second-level early warning of link disturbance is realized, and data congestion caused by sudden quality reduction is avoided; a dynamic discrimination mechanism driven by a precursor score is utilized, so that a scheduling decision is converted from passive response to active adjustment, and the probability of cascade failure caused by mass mutation is greatly reduced; an interference spectrum adaptive strategy enables the system to adaptively match the optimal coding mode and the redundant structure when facing narrow-band noise, periodic interference or transient impact, so that the anti-interference capability and the bandwidth utilization efficiency are improved; priority adjustment is carried out in combination with a link risk trend and a state evolution rule, and the protection timeliness and resource allocation efficiency of the key link are remarkably improved.

The invention discloses a computer network low-power-consumption communication optimization regulation and control method adaptive to a complex environment. Multi-dimensional environment parameters are collected and preprocessed through a multi-dimensional environment parameter sensing module; training is carried out through combination of multi-modal feature interaction and a gradient boosting tree algorithm, a low-power-consumption communication optimization model is established, and a related threshold value is output; distributing network resources through a hierarchical constraint and velocity attenuation fused particle swarm optimization algorithm; the communication state is monitored in real time, and when it is monitored that the communication energy consumption exceeds the energy consumption reference threshold value or the communication quality is lower than the minimum quality guarantee threshold value, feedback adjustment is triggered and resources are redistributed; model input parameters are dynamically updated, a parameter mapping library is established, and when the network environment changes, the model is finely adjusted through the parameter mapping library and a new scene is adapted. According to the invention, by realizing the balance between the sensing precision and the sampling energy consumption, the prediction accuracy, the bandwidth utilization rate and the scene adaptation efficiency are improved, and the low-power-consumption and high-quality communication of the network in a complex environment is ensured.

The invention discloses a topology awareness-based multi-core hybrid cluster resource scheduling method, which comprises the following steps of: obtaining a host index and DMA (Direct Memory Access) data, and generating a PCIe (Peripheral Component Interconnect Express) topological graph, a transmission mode baseline and a bandwidth competition matrix; based on the topological graph and the base line, segmenting large DMA transmission into micro-packets for staggered scheduling, and constructing a predictive recombination cache; detecting an avalanche effect by monitoring a descending gradient of prediction accuracy, when the avalanche effect is detected, generating a deterministic space-time braided matrix based on a bandwidth competition matrix, and exporting a mode switching strategy; and executing adaptive dual-mode scheduling of a prediction mode and a deterministic mode according to the strategy. Through fine-grained micro-fragment scheduling and a robust dual-mode switching mechanism, the transmission delay is remarkably reduced, and the PCIe bandwidth utilization rate and the system stability are improved.

The invention provides a data transmission method, system and device, electronic equipment and a storage medium, and the method comprises the steps: determining a target cache queue from a plurality of cache queues according to the state information of a data sender and the state information of a first cache queue in the plurality of cache queues when obtaining new to-be-sent data, the state information of the data sender comprises a software system state and / or a bus state, writing new to-be-sent data into a target cache queue, merging the data in the target cache queue into a merged data packet, mounting the merged data packet to a preset first-in first-out queue, and sending the merged data packet to a data receiver, and sending a plurality of data obtained by analyzing the merged data packet to the corresponding upper-layer application by the data receiver. According to the method, the buffer queue structure is additionally arranged, the small data packets are merged into the large data packets to be sent, the transmission interaction frequency of PCIE is reduced, the bandwidth utilization rate is increased, and CPU resource waste is reduced.

The invention discloses a reasoning optimization method of a pre-training model and electronic equipment, and relates to the technical field of reasoning of the pre-training model.The priority of each subtask is determined on the basis of load data of multiple task stages of the pre-training model.The calculation units are allocated to the subtasks according to the priorities of the subtasks, and therefore the calculation efficiency of the subtasks is improved. By dynamically decoupling each task stage, efficient allocation of computing resources is realized, and the resource utilization rate is improved. And on the other hand, the first data of the two adjacent task stages are transmitted through a double-buffering mechanism, the problem that the inter-layer data transmission efficiency is low in the related technology is solved, and the hardware bandwidth utilization rate is increased. Therefore, the technical problems of low utilization rate of idle memory bandwidth resources and low interlayer transmission efficiency are solved, and the technical effect of improving the resource utilization rate and the bandwidth utilization rate is achieved.