4results about How to "Improve execution time" patented technology

The application provides a configuration method of a test task of a test interface and a test method of a device, which comprises the following steps: setting a test action capable of being executed by a test device in an action setting window to form an action set; setting state information capable of being fed back by a device under test in a judgment setting window to form a state set; setting at least one test flow in a test flow window; clicking the test flow in the test flow window; and configuring test step information corresponding to the test flow based on the action set and the state set in a test step window to form an association relationship between the test flow and the corresponding test step information, wherein the test step information comprises test actions selected from the action set, state information selected from the state set, and the order of the test actions and the state information; and generating a test task based on the association relationship between the test flow and the corresponding test step information, so as to solve the problem of high workload of test script development.

The application provides a reconfigurable computing architecture mapping method based on graph convolution and reinforcement learning, comprising the following steps: acquiring a data dependency graph and a coarse-grained reconfigurable array, and calculating a minimum start interval of the data dependency graph; preprocessing the data dependency graph according to the minimum start interval; simplifying a hardware model resource graph according to the structure of the coarse-grained reconfigurable array; taking a node set of the hardware model resource graph as an action space; adopting a reinforcement learning model based on a graph convolution network to map the preprocessed data dependency graph, so as to obtain a mapping result of a data dependency graph node on the hardware model resource graph; calculating a reward value of the mapping result; and when it is judged according to the reward value that the mapping result meets a successful mapping condition, outputting the mapping result. The application optimizes the layout and wiring of a loop on a reconfigurable computing array by using graph convolution and reinforcement learning, improves the parallelism of loop execution, and thus increases the mapping quality and reduces the execution time of image denoising.

The present application relates to the field of resource optimization, and more particularly to a LoRA fine-tuning computing resource dynamic allocation method, by obtaining the video memory state and computing unit load data stream of the computing cluster, the prior features representing the resource fragmentation degree and load level are obtained according to the historical time series data analysis, and the bottleneck label is set for various computing tasks accordingly; during task execution, differentiated scheduling optimization is carried out according to the label type and real-time resource features, the video memory bottleneck task is temporarily withheld until it is reduced to the predetermined threshold value before execution by monitoring the real-time video memory state features; the computing bottleneck task is started when the load meets the conditions by monitoring the real-time load state features. The present application realizes the perception response of the dynamic resource state of the cluster and the targeted scheduling of the task bottleneck type, alleviates the congestion and idling caused by video memory fragmentation and uneven resource distribution, and improves the resource utilization balance and the overall execution efficiency of the LoRA fine-tuning task.

This invention discloses a customer group operation method, apparatus, electronic device, and storage medium, relating to the field of internet data processing technology. The method includes: dynamically setting the number of shards and grouping user data into multiple independently executable micro-batch task units by assessing cluster size and resource status. A streaming engine with multiple execution nodes is used to adjust the micro-batch granularity in real time according to cluster load. Each execution node monitors resource status in real time, calculates a resource busyness coefficient based on the current number of micro-batch task units and estimated execution time, and dynamically allocates task volume accordingly. In the face of upstream tag data latency, the system automatically triggers a tiered degradation strategy, flexibly cutting non-core tasks, switching to offline tag snapshots, or implementing a fallback strategy based on the latency level and task priority, ensuring the continuity of core business. This invention effectively solves the problems of poor timeliness, low resource utilization, and lack of effective disaster recovery mechanisms in existing technologies.