Allocating resources for a machine learning model

1. A computer-implemented method for allocating resources for a machine learning model, the method comprising:

obtaining a data flow graph representative of a machine learning model to execute on a special purpose machine learning model processor;

scheduling operations of the data flow graph representative of the machine learning model, at least by determining: i) a starting clock cycle and ii) a duration for execution of each operation of the data flow graph, based on a deterministic attribute specific to computational operations of the data flow graph;

compiling the data flow graph using a deterministic instruction set architecture that specifies functionality of the special purpose machine learning model processor;

determining an amount of resources required to execute the data flow graph representative of the machine learning model; and

allocating resources based on the determined amounts of resources required to execute the machine learning model.

2. The computer-implemented method of claim 1 , wherein the data flow graph comprises a plurality of nodes, a plurality of connector directed edges, and a plurality of parameter directed edges, wherein each node represents a respective operation performed by the machine learning model, wherein each connector directed edge connects a respective first node to a respective second node that represents an operation that receives, as input, an output of an operation represented by the respective first node, and wherein each parameter directed edge connects into a respective node and represents a flow of one or more parameters of the machine learning model as input to the operation represented by the respective node.

3. The computer-implemented method of claim 1 , wherein determining an amount of resources required to execute the data flow graph comprises:

determining a number of operations required for executing the data flow graph;

determining an amount of storage required for executing the data flow graph; and

determining an amount of input/output communications required for executing the data flow graph.

4. The computer-implemented method of claim 1 , further comprising automatically load-balancing the machine learning model based on operations, input, and output required of the machine learning model at compile time.

5. The computer-implemented method of claim 1 , wherein the machine learning model executes on the special purpose machine learning model processor in a data center.

6. The computer-implemented method of claim 5 , further comprising adjusting data center resources based on resource use of the machine learning model.

7. The computer-implemented method of claim 1 , wherein allocating resources based on the determined amounts of resources required to execute the machine learning model comprises:

identifying resources for a second machine learning model that executes on the special purpose machine learning model processor, the resources being unused during an allocated time period; and

providing the resources to the machine learning model represented by the data flow graph during the allocated time period until a request for the second machine learning model is received by the special purpose machine learning model processor.

8. The computer-implemented method of claim 1 , further comprising:

rebalancing the machine learning model based on factors including usage, timing, number of operations the machine learning model requires, or a precision necessary to perform a machine learning task.

9. A system comprising one or more computers and one or more storage devices storing instructions that are executable by the one or more computers to cause the one or more computers to perform operations of:

obtaining a data flow graph representative of a machine learning model to execute on a special purpose machine learning model processor;

scheduling operations of the data flow graph representative of the machine learning model, at least by determining: i) a starting clock cycle and ii) a duration for execution of each operation of the data flow graph, based on a deterministic attribute specific to computational operations of the data flow graph;

compiling the data flow graph using a deterministic instruction set architecture that specifies functionality of the special purpose machine learning model processor;

determining an amount of resources required to execute the data flow graph representative of the machine learning model; and

allocating resources based on the determined amounts of resources required to execute the machine learning model.

10. The system of claim 9 , wherein the data flow graph comprises a plurality of nodes, a plurality of connector directed edges, and a plurality of parameter directed edges, wherein each node represents a respective operation performed by the machine learning model, wherein each connector directed edge connects a respective first node to a respective second node that represents an operation that receives, as input, an output of an operation represented by the respective first node, and wherein each parameter directed edge connects into a respective node and represents a flow of one or more parameters of the model as input to the operation represented by the respective node.

11. The system of claim 9 , wherein determining an amount of resources required to execute the data flow graph comprises:

determining a number of operations required for executing the data flow graph;

determining an amount of storage required for executing the data flow graph; and

determining an amount of input/output communications required for executing the data flow graph.

12. The system of claim 9 , wherein the operations comprise automatically load-balancing the machine learning model based on operations and input/output required of the machine learning model at compile time.

13. The system of claim 9 , wherein the special purpose machine learning model processor is in a data center.

14. The system of claim 13 , wherein the operations comprise adjusting data center resources based on resource use of the machine learning model.

15. The system of claim 9 , wherein allocating resources based on the determined amounts of resources required to execute the machine learning model comprises:

identifying resources for a second machine learning model that executes on the special purpose machine learning model processor, the resources being unused during an allocated time period; and

providing the resources to the machine learning model during the allocated time period until a request for the second machine learning model is received by the special purpose machine learning model processor.

16. The system of claim 9 , wherein the operations comprise rebalancing the machine learning model based on factors including usage, timing, number of operations the model requires, or a precision necessary to perform a machine learning task.

17. A computer-implemented method for allocating resources for a machine learning model, the method comprising:

obtaining a data flow graph representative of a machine learning model to execute on a special purpose machine learning model processor,

wherein the data flow graph comprises a plurality of nodes, a plurality of connector directed edges, and a plurality of parameter directed edges, wherein each node represents a respective operation performed by the machine learning model, wherein each connector directed edge connects a respective first node to a respective second node that represents an operation that receives, as input, an output of an operation represented by the respective first node, and wherein each parameter directed edge connects into a respective node and represents a flow of one or more parameters of the machine learning model as input to the operation represented by the respective node;

scheduling operations of the data flow graph representative of the machine learning model,

wherein scheduling operations of the data flow graph comprises:

determining a starting clock cycle,

determining a duration for execution of each operation, and

scheduling the operations of the data flow graph using the starting clock cycle and the duration;

compiling the data flow graph using a deterministic instruction set architecture that specifies functionality of the special purpose machine learning model processor;

determining an amount of resources required to execute the data flow graph representative of the machine learning model; and

allocating resources based on the determined amounts of resources required to execute the machine learning model.

18. A system comprising one or more computers and one or more storage devices storing instructions that are executable by the one or more computers to cause the one or more computers to perform operations of:

obtaining a data flow graph representative of a machine learning model to execute on a special purpose machine learning model processor,

wherein the data flow graph comprises a plurality of nodes, a plurality of connector directed edges, and a plurality of parameter directed edges, wherein each node represents a respective operation performed by the machine learning model, wherein each connector directed edge connects a respective first node to a respective second node that represents an operation that receives, as input, an output of an operation represented by the respective first node, and wherein each parameter directed edge connects into a respective node and represents a flow of one or more parameters of the machine learning model as input to the operation represented by the respective node;

scheduling operations of the data flow graph, wherein scheduling operations of the data flow graph comprises:

determining a starting clock cycle,

determining a duration for execution of each operation, and

scheduling the operations of the data flow graph using the starting clock cycle and the duration;

compiling the data flow graph using a deterministic instruction set architecture that specifies functionality of the special purpose machine learning model processor;

determining an amount of resources required to execute the data flow graph representative of the machine learning model; and

allocating resources based on the determined amounts of resources required to execute the machine learning model.