Back to results
US 11,699,073 B2Granted
Network off-line model processing method, artificial intelligence processing device and related products
AssigneeCambricon Technologies Corporation Limited
InventorsWeiguang Kong, Yaling Huang, Jin Wang
Life of this patent
Estimating the term…
Abstract
The present disclosure provides a network off-line model processing method, an artificial intelligence processing device and related products, where the related products include a combined processing device. The combined processing device includes the artificial intelligence processing device, a general-purpose interconnection interface, and other processing devices, where the artificial intelligence processing device interacts with the other processing devices to jointly complete computation designated by users. The embodiments of the present disclosure can accelerate the operation of the network off-line model.
Claims8
A processing method of a network off-line model, comprising:
- obtaining information of an operating unit of each sub-network in the network off-line model, wherein the information of the operating unit includes a correspondence between the sub-network and a type of the operating unit, and the type of the operating unit includes a general-purpose processing unit type or an artificial intelligence (AI) processing unit type;
- defining operating parameters of the sub-network according to the information of the operating unit when the network off-line model is being constructed to obtain a constructed network off-line model, wherein the operating parameters of the sub-network are used to represent the type of the operating unit of each sub-network
- operating the constructed network off-line model, which includes:
- determining the operating unit corresponding to a target sub-network according to the operating parameters of the sub-network, wherein the target sub-network is any sub-network of the network off-line model;
- and
- operating the target sub-network on the corresponding operating unit to operate the network off-line model
- wherein if the operating unit corresponding to the target sub-network is the AI processing unit, the operating the target sub-network on the corresponding operating unit to operate the network off-line model includes:
- when an underlying library interface is being called, obtaining a channel identifier transmitted from the underlying liability interface;
- determining a channel of the AI processing unit for transmitting data according to the channel identifier;
- and
- operating the target sub-network on the AI processing unit through the channel to operate the network off-line model.
8 more elements
An artificial intelligence (AI) processing device, comprising:
- an obtaining module configured to obtain information of an operating unit of each sub-network in a network off-line model, where the information of the operating unit includes a correspondence between the sub-network and a type of the operating unit, and the type of the operating unit includes a general-purpose processing unit type or an AI processing unit type;
- a constructing module configured to define operating parameters of the sub-network according to the information of the operating unit when the network off-line model is being constructed to obtain a constructed network off-line model, wherein the operating parameters of the sub-network are used to represent the type of the operating unit of each sub-network;
- an operating module configured to operate the constructed network off-line model, wherein the operating module is configured to:
- determine the operating unit corresponding to a target sub-network according to the operating parameters of the sub-network, where the target sub-network is any sub-network of the network off-line model;
- and
- operate the target sub-network in the corresponding operating unit to operate the network off-line model,
- wherein if the operating unit corresponding to the target sub-network is an AI processing unit, the operating module is configured to operate the target sub-network on the corresponding operating unit to operate the network off-line model, wherein the operating module is configured to:
- when an underlying library interface is being called, obtain a channel identifier transmitted from the underlying liability interface;
- determine a channel of the AI processing unit for transmitting data according to the channel identifier;
- and
- operate the target sub-network on the AI processing unit through the channel to operate the network off-line model.
8 more elements
Dependencies are inferred from the claim text.