Artificial intelligence computing resource allocation system with heterogeneous computing architecture

Abstract

An AI computing resource allocation system with a heterogeneous computing architecture includes a user device and a server. The user device generates a requisition request and detects a voice message. The server connects to the user device to receive the requisition request and the voice message. The server allocates a secondary computing unit and a primary computing unit to the user device. The secondary computing unit generates an identification text based on the voice message. The primary computing unit receives the identification text and generates a response text based on the identification text, which is then sent to the secondary computing unit. When the secondary computing unit receives the response text, it generates a response voice message and a digital humanoid image based on the response text and sends them to the user device.

Claims

1. An artificial intelligence computing resource allocation system with a heterogeneous computing architecture, comprising: a user device for generating a requisition request and detecting a voice message; and a server connected to the user device to receive the requisition request and the voice message, the server allocating a secondary computing unit and a primary computing unit to the user device, the secondary computing unit generating a recognized text based on the voice message, and the primary computing unit receiving the recognized text and generating a response text to the secondary computing unit based on the recognized text; wherein... When the secondary processing unit receives the response text, it generates a response voice message and a digital humanoid image based on the response text and sends them to the user device, causing the user device to play the response voice message and the digital humanoid image.

2. The AI ​​computing resource allocation system with a heterogeneous computing architecture as described in claim 1, the server further comprising: a gateway for receiving the requisition request; a heterogeneous computing cluster connected to the gateway, wherein the heterogeneous computing cluster comprises multiple secondary computing units and multiple primary computing units; wherein, When the gateway receives the requisition request, the gateway dynamically schedules one of the multiple secondary computing units and one of the multiple primary computing units to establish a connection between the secondary computing unit and the primary computing unit and the user device.

3. The artificial intelligence computing resource allocation system with a heterogeneous computing architecture as described in claim 2, wherein: The heterogeneous computing cluster includes a relational database used to store information related to the user device; when the secondary computing unit and the primary computing unit lose connection with the user device, the server clears the information in the secondary computing unit, the primary computing unit, and the relational database.

4. An artificial intelligence computing resource allocation system with a heterogeneous computing architecture as described in claim 1, wherein: The secondary processing unit is a high-memory bandwidth image processor. The secondary processing unit executes program data of a speech synthesis model to generate the response speech message based on the response text and executes program data of an image rendering model to generate the digital humanoid image.

5. An artificial intelligence computing resource allocation system with a heterogeneous computing architecture as described in claim 4, wherein: The secondary computing unit executes program data of an automatic speech recognition model to generate the recognized text based on the speech message.

6. The AI ​​computing resource allocation system with a heterogeneous computing architecture as described in claim 1, wherein the main computing unit is a processor with a unified memory architecture and multiple cores, the main computing unit executes program data of an AI platform and a large language model; the AI ​​platform decouples the underlying model through a standardized application compiler interface and has retrieval-enhanced generation capabilities to convert the identified text into at least one vector and retrieve multiple text fragments; the large language model receives the multiple text fragments to infer and generate the response text.

7. An artificial intelligence computing resource allocation system with a heterogeneous computing architecture as described in claim 6, wherein: The main computing unit includes a vector database for storing the at least one vector.

8. An artificial intelligence computing resource allocation system with a heterogeneous computing architecture as described in claim 1, wherein: The user device is connected to the server via a closed local area network.

Images