Patents
Literature
Patsnap Eureka AI that helps you search prior art, draft patents, and assess FTO risks, powered by patent and scientific literature data.

2results about How to "Implement parallel execution" patented technology

Message forwarding method and device, computer device, readable storage medium and program product

ActiveCN119484469BImplement parallel executionImprove forwarding efficiencyTransmissionMessage processingTerm memory
The application relates to a message forwarding method and device, computer equipment, a readable storage medium and a program product, and relates to the field of message processing. The method comprises the following steps: when a first object sends a request message to a second object, source address information of the request message is acquired; the source address information is subjected to hash processing to obtain a first hash value of the source address information; a first memory core matched with the first hash value is screened out from a plurality of first candidate memory cores used for message processing of the request message; the request message is subjected to message processing through the first memory core to obtain a first target message, and the first target message is forwarded to the second object. The method can improve the message forwarding efficiency.
Owner:CHINA TELECOM CLOUD TECH CO LTD

Chip calculation reasoning method and system of hybrid expert model

ActiveCN121920538AReduce loading delayImprove cache hit ratioBiological modelsInference methodsEngineeringTerm memory
The invention provides a chip calculation reasoning method and system for a hybrid expert model, and relates to the technical field of artificial intelligence calculation, and the method comprises the steps: obtaining input data, calculating an expert activation weight, carrying out the time sequence tracking, carrying out the clustering analysis, recognizing an expert combination, and preloading the weight to a shared cache; selecting a target expert based on the activation weight, and loading the weight; fragmenting the data, constructing an assembly line level, and mapping the data to different computing units; and each unit is controlled to execute calculation and weighted summation according to an assembly line to obtain a result. Through expert activation mode prediction and pipeline parallel processing, the reasoning efficiency is improved, and the memory access overhead is reduced.
Owner:XINQIAO (BEIJING) SEMICONDUCTOR CO LTD