Method, device and computer program product for determining resource amount for dedicated processing resources
Embodiments of the present disclosure relate to a method, device and computer program product for determining a resource amount of dedicated processing resources. The method comprises obtaining a structural representation of a neural network for deep learning processing, the structural representation indicating a layer attribute of the neural network that is associated with the dedicated processing resources; and determining the resource amount of the dedicated processing resources required for the deep learning processing based on the structural representation. In this manner, the resource amount of the dedicated processing resources required by the deep learning processing may be better estimated to improve the performance and resource utilization rate of the dedicated processing resource scheduling.
1. A method of determining a resource amount of dedicated processing resources, comprising:
obtaining a structural representation of a neural network for deep learning processing, the structural representation indicating a layer attribute of the neural network that is associated with the dedicated processing resources; and
determining the resource amount of the dedicated processing resources required for the deep learning processing based on the structural representation.
2. The method according to claim 1 , wherein obtaining the structural representation comprises:
obtaining a specific structural representation of the neural network, the specific structural representation having a specific form of representation for a deep learning application performing the deep learning processing; and
normalizing the specific structural representation as the structural representation.
3. The method according to claim 2 , wherein normalizing the specific structural representation as the structural representation comprises:
determining a layer attribute associated with a neural network layer of the neural network based on the specific structural representation; and
normalizing the specific structural representation as the structural representation based on the layer attribute.
4. The method according to claim 3 , wherein determining the layer attribute comprises determining at least one of:
an identifier of the neural network layer;
a type of the neural network layer;
an upstream neural network layer of the neural network layer;
a downstream neural network layer of the neural network layer; and
a configurable attribute of the neural network layer.
5. The method according to claim 1 , wherein obtaining the structural representation comprises:
obtaining a file containing the structural representation; and
parsing the file to obtain the structural representation.
6. The method according to claim 1 , wherein obtaining the structural representation comprises:
requesting the structural representation from a deep learning application performing the deep learning processing; and
in response to receiving a response from the deep learning application, obtaining the structural representation from the response.
7. The method according to claim 1 , wherein determining the resource amount comprises:
determining at least one of a memory resource amount and a computing resource amount of the dedicated processing resources required for the deep learning processing.
8. The method according to claim 7 , wherein determining the memory resource amount comprises:
determining a layer memory resource amount for each neural network layer of the neural network based on the structural representation, the layer memory resource amount indicating a memory of the dedicated processing resources required for a corresponding neural network layer; and
determining the memory resource amount based on a sum of the layer memory resource amounts.
9. The method according to claim 7 , wherein determining the computing resource amount comprises:
determining a layer computing resource amount for each neural network layer of the neural network based on the structural representation, the layer computing resource amount indicating a computing capability of the dedicated processing resources required for a corresponding neural network layer;
selecting, from the layer computing resource amounts, a target layer computing resource amount requiring a computing capability above a predetermined threshold; and
determining the computing resource amount based on the target layer computing resource amount.
10. A device for determining a resource amount of dedicated processing resources, comprising:
at least one processing unit; and
at least one memory coupled to the at least one processing unit and storing instructions executed by the at least one processing unit, the instructions, when executed by the at least one processing unit, causing the device to implement acts, comprising:
obtaining a structural representation of a neural network for deep learning processing, the structural representation indicating a layer attribute of the neural network that is associated with the dedicated processing resources; and
determining the resource amount of the dedicated processing resources required for the deep learning processing based on the structural representation.
11. The device according to claim 10 , wherein obtaining the structural representation comprises:
obtaining a specific structural representation of the neural network, the specific structural representation having a specific form of representation for a deep learning application performing the deep learning processing; and
normalizing the specific structural representation as the structural representation.
12. The device according to claim 11 , wherein normalizing the specific structural representation as the structural representation comprises:
determining a layer attribute associated with a neural network layer of the neural network based on the specific structural representation; and
normalizing the specific structural representation as the structural representation based on the layer attribute.
13. The device according to claim 12 , wherein determining the layer attribute comprises determining at least one of:
an identifier of the neural network layer;
a type of the neural network layer;
an upstream neural network layer of the neural network layer;
a downstream neural network layer of the neural network layer; and
a configurable attribute of the neural network layer.
14. The device according to claim 10 , wherein obtaining the structural representation comprises:
obtaining a file containing the structural representation; and
parsing the file to obtain the structural representation.
15. The device according to claim 10 , wherein obtaining the structural representation comprises:
requesting the structural representation from a deep learning application performing the deep learning processing; and
in response to receiving a response from the deep learning application, obtaining the structural representation from the response.
16. The device according to claim 10 , wherein determining the resource amount comprises:
determining at least one of a memory resource amount and a computing resource amount of the dedicated processing resources required for the deep learning processing.
17. The device according to claim 16 , wherein determining the memory resource amount comprises:
determining a layer memory resource amount for each neural network layer of the neural network based on the structural representation, the layer memory resource amount indicating a memory of the dedicated processing resources required for a corresponding neural network layer; and
determining the memory resource amount based on a sum of the layer memory resource amounts.
18. The device according to claim 16 , wherein determining the computing resource amount comprises:
determining a layer computing resource amount for each neural network layer of the neural network based on the structural representation, the layer computing resource amount indicating a computing capability of the dedicated processing resources required for a corresponding neural network layer;
selecting, from the layer computing resource amounts, a target layer computing resource amount requiring a computing capability above a predetermined threshold; and
determining the computing resource amount based on the target layer computing resource amount.
19. A computer program product being tangibly stored on a non-transient computer readable medium and comprising machine executable instructions, the machine executable instructions, when executed, cause a machine to perform steps of a method of determining a resource amount of dedicated processing resources, comprising:
obtaining a structural representation of a neural network for deep learning processing, the structural representation indicating a layer attribute of the neural network that is associated with the dedicated processing resources; and
determining the resource amount of the dedicated processing resources required for the deep learning processing based on the structural representation.