Shanghai Cambricon Information Technology Co., Ltd

Cambricon Technologies Corporation Limited

Cambricon (xi'an) Semiconductor Co., Ltd.

Xunyu Chen

Linyang Wu

Weijian Du

Qi Guo

Tianshi Chen

Yunji Chen

Kangyu Wang

Tian Zhi

Jie Wei

Xiaofu Meng

Zhenxing Zhang

A neural network processing method, comprising the following steps: obtaining a model dataset and model structure parameters of an original network (S); obtaining an operational attribute of each compute node in the original network; operating the original network according to the model dataset and the model structure parameters of the original network and the operational attribute of each compute node, to obtain an instruction corresponding to each compute node in the original network (S); and if the operational attribute of the current compute node is a first operational attribute, storing a network weight and the instruction corresponding to the current compute node into a first non-volatile memory, so as to obtain a first offline model corresponding to the original network (S). Further provided are a computer system and a storage medium. The neural network processing method, the computer system, and the storage medium shorten the time for a processor to operate the same network, and improve the processing speed and efficiency of the processor.

Neural network processing method, computer system and storage medium

The present disclosure relates to a method, a device and a computation apparatus for allocating a space address to data in a memory, where the computation apparatus is included in a combined processing apparatus, which includes a general interconnection interface and other processing apparatuses. The computation apparatus interacts with other processing apparatuses to jointly complete computations specified by the user. The combined processing apparatus also includes a storage apparatus. The storage apparatus is respectively connected to the computation apparatus and the other processing apparatuses, and is used for storing data of the computation apparatus and other processing apparatuses. The technical solutions of the present disclosure improve utilization of storage space of the memory.

Method and device for allocating storage addresses for data in memory

Aspects for Long Short-Term Memory (LSTM) blocks in a recurrent neural network (RNN) are described herein. As an example, the aspects may include one or more slave computation modules, an interconnection unit, and a master computation module collectively configured to calculate an activated input gate value, an activated forget gate value, a current cell status of the current computation period, an activated output gate value, and a forward pass result.

Apparatus and method for executing recurrent neural network and LSTM computations

Disclosed are a general machine learning model generation method and apparatus, and a computer device and a storage medium. The method comprises: acquiring task parameters of a machine learning task (S); performing classification processing on the task parameters to obtain task instructions and model parameters (S); aggregating the task instructions and the model parameters according to a data type to obtain stack data and heap data (S); and integrating the stack data and the heap data to obtain a general machine learning model (S). By means of the method, compiled results of a corresponding general model in the running of an algorithm can be directly executed, which avoids repetitive compilation, thus greatly improving the efficiency of machine learning algorithm implementation and shortening the time from compilation to obtaining execution results.

General machine learning model, and model file generation and parsing method

Disclosed are a general-purpose machine learning model generation method and apparatus, and a computer device and a storage medium. The method comprises: acquiring task parameters of a machine learning task (S), performing classification processing on the task parameters to obtain task instructions and model parameters (S), aggregating the task instructions and the model parameters according to a data type to obtain stack data and heap data (S), and integrating the stack data and the heap data to obtain a general-purpose machine learning model (S). By means of the method, compiled results of a corresponding general-purpose model in the running of an algorithm can be directly executed, which avoids repetitive compilation, thus greatly improving the efficiency of machine learning algorithm implementation and shortening the time from compilation to obtaining execution results.

Computer systems, data processing methods, and computer-readable media are provided to run original networks. An exemplary computer system includes first and second processors a memory storing offline models and corresponding input data of a plurality of original networks, and a runtime system configured to run on the first processor. The runtime system, when runs on the first processor, causes the first processor to implement a plurality of virtual devices comprising a data processing device configured to obtain an offline model and corresponding input data of an original network from the memory, an equipment management device configured to control turning on or off of the second processor, and a task execution device configured to control the second processor to run the offline model of the original network.

Task parallel processing method, apparatus and system, storage medium and computer device

Computer systems, data processing methods, and computer-readable media are provided to run original networks. An exemplary computer system includes first and second processors and first and second memories. The first memory stores offline models and corresponding input data of a plurality of original networks, and a runtime system configured to run on the first processor. The second memory stores an operating system configured to run on the first processor or the second processor. When the runtime system runs on the first processor, the runtime system obtains an offline model and corresponding input data of an original network from the first memory and controls the second processor to run the offline model of the original network. The offline model of the original network includes model parameters, instructions, and interface data of respective computation nodes of the original network.

Growing community of inventors

Pudong New Area, China

Xunyu Chen