公开(公告)号：US12223300B2

公开(公告)日：2025-02-11

申请号：US18309341

申请日：2023-04-28

Applicant: MEDIATEK INC.

Inventor： Meng-Hsuan Yang ,  Po-hua Huang ,  Hsing-Chang Chou ,  Ting Chen Tsan ,  Yu-Lung Lu

IPC: G06F8/41 ,  G06N3/10

Abstract: A method of compiling a deep learning model includes reading metadata from a compiled result, the metadata indicating a structure of the deep learning model corresponding to a low-level IR, receiving shape information of an input tensor of the deep learning model, determining a shape of an output tensor of a first computation operation of the computation operations based on the shape information of the input tensor of the deep learning model and the structure of the deep learning model, tiling the output tensor of the first computation operation into one or more tiles according to the shape of the output tensor of the first computation operation and hardware limitations of a processor executing the deep learning model, and patching one or more copies of a templated hardware command into executable hardware commands.

公开(公告)号：US20250036977A1

公开(公告)日：2025-01-30

申请号：US18751335

申请日：2024-06-23

Applicant: MEDIATEK INC.

Inventor： Chia-Wei Hsu ,  Yu-Lung Lu ,  Yen-Ting Chiang ,  Chih-wei Chen ,  Yi-Cheng Lu ,  Jia-Sian Hong ,  Kuan-Yu Chen ,  Pei-Kuei Tsung ,  Hua Wu

IPC: G06N5/04

Abstract: An electronic device is configured to execute instructions: compiling a first AI model and second AI model(s) to a first compiled file and second compiled file(s), respectively, wherein the first compiled file comprises a first data set and a first command set, and the second compiled file(s) comprises second data set(s) and second command set(s); generating light version file(s) for the AI model(s), wherein the light version file(s) comprises the second command set(s) and data patch(es); storing the first compiled file and the light version file(s) to a storage device; loading the first compiled file from the storage device to a memory; loading the light version file(s) from the storage device to the memory; generating the second data set(s) according to the first data set and the data patch(es); and executing the second AI model(s) according to the generated second data set(s) and the second command set(s) in the memory.

公开(公告)号：US20170269950A1

公开(公告)日：2017-09-21

申请号：US15070424

申请日：2016-03-15

Applicant: MediaTek Inc.

Inventor： Yu-Lung Lu ,  Hong-Rong Hsu

IPC: G06F9/455 ,  G06F9/45

CPC classification number: G06F9/45508 ,  G06F8/41 ,  G06F9/44547 ,  G06F9/4552

Abstract: A device selectively executes native machine code of a computing method in an application. Prior to execution of the application, a predicted usage level of the computing method is determined based on available statistical analysis data of the computing method. According to a determination of whether the predicted usage level exceeds a threshold, a selector selects executable code of the computing method for execution. The executable code is the native machine code or bytecode of the computing method. When the computing method is called during execution of the application, the selected executable code is loaded from non-volatile storage into memory for execution by a virtual machine. Furthermore, runtime usage level of the computing method is monitored to determine whether to switch from bytecode to native machine code execution.