Flat representation of machine learning model
The example embodiments are directed to a system and method for deploying a machine learning model using a parse-free memory allocation. In one example, the method may include one or more of receiving a request to deploy a machine learning model, in response to receiving the request, creating a memory map comprising a mapping of a data structure for storing an unpacked flat representation of the machine learning model, allocating a contiguous block of memory of the data structure that is mapped by the memory map, loading data blocks of the unpacked flat representation of the machine learning model into the allocated contiguous blocks of memory of the data structure, and storing an offset associated with the contiguous block of memory in storage.
1. A computing system comprising:
a storage; and
a processor in communication with the storage and configured to:
receive a request to deploy a machine learning model,
in response to receiving the request, create a memory map comprising a mapping of a data structure for storing an unpacked flat representation of the machine learning model,
allocate a contiguous block of memory of the data structure mapped by the memory map,
load data blocks of the unpacked flat representation of the machine learning model into the allocated contiguous block of memory of the data structure, and
store an offset associated with the contiguous block of memory in the storage.
2. The computing system of claim 1 , wherein the unpacked flat representation of the machine learning model comprises a parse-free memory model which can be traversed without parsing and unpacking.
3. The computing system of claim 1 , wherein the data structure storing the unpacked flat representation of the machine learning model comprises a data file having an organizational structure mapped by the memory map.
4. The computing system of claim 1 , wherein the unpacked flat representation of the machine learning model comprises a tree structure of objects, and the objects are connected via offsets.
5. The computing system of claim 4 , wherein each object comprises fields including one or more of a type code, a length, and an offset that points to a list of values.
6. The computing system of claim 1 , wherein the processor is configured to initially load metadata of the unpacked flat representation of the machine learning model into the data structure, and then load the data blocks of the unpacked flat representation of the machine learning model into the data structure.
7. The computing system of claim 6 , wherein the metadata is loaded from a root object of the unpacked flat representation of the machine learning model.
8. The computing system of claim 1 , wherein the processor is further configured to execute an application that launches the unpacked flat representation of the machine learning model.
9. A computer-implemented method comprising:
receiving a request to deploy a machine learning model;
in response to receiving the request, creating a memory map comprising a mapping of a data structure for storing an unpacked flat representation of the machine learning model;
allocating a contiguous block of memory of the data structure that is mapped by the memory map;
loading data blocks of the unpacked flat representation of the machine learning model into the allocated contiguous block of memory of the data structure; and
storing an offset associated with the contiguous block of memory in a storage.
10. The computer-implemented method of claim 9 , wherein the unpacked flat representation of the machine learning model comprises a parse-free memory model which can be traversed without parsing and unpacking.
11. The computer-implemented method of claim 9 , wherein the data structure storing the unpacked flat representation of the machine learning model comprises a data file having an organizational structure mapped by the memory map.
12. The computer-implemented method of claim 9 , wherein the unpacked flat representation of the machine learning model comprises a tree structure of objects, and the objects are connected via offsets.
13. The computer-implemented method of claim 12 , wherein each object comprises fields including one or more of a type code, a length, and an offset that points to a list of values.
14. The computer-implemented method of claim 9 , wherein the loading comprises initially loading metadata of the unpacked flat representation of the machine learning model into the data structure, and then loading the data blocks of the unpacked flat representation of the machine learning model into the data structure.
15. The computer-implemented method of claim 14 , wherein the metadata is loaded from a root object of the unpacked flat representation of the machine learning model.
16. The computer-implemented method of claim 9 , further comprising executing an application that launches the unpacked flat representation of the machine learning model.
17. A non-transitory computer readable medium comprising program instructions which when executed cause a processor to perform a method comprising:
receiving a request to deploy a machine learning model;
in response to receiving the request, creating a memory map comprising a mapping of a data structure for storing an unpacked flat representation of the machine learning model;
allocating a contiguous block of memory of the data structure that is mapped by the memory map;
loading data blocks of the unpacked flat representation of the machine learning model into the allocated contiguous block of memory of the data structure; and
storing an offset associated with the contiguous block of memory in a storage.
18. The non-transitory computer readable medium of claim 17 , wherein the unpacked flat representation of the machine learning model comprises a parse-free memory model which can be traversed without parsing and unpacking.
19. The non-transitory computer readable medium of claim 17 , wherein the data structure storing the unpacked flat representation of the machine learning model comprises a data file having an organizational structure mapped by the memory map.
20. The non-transitory computer readable medium of claim 17 , wherein the unpacked flat representation of the machine learning model comprises a tree structure of objects, and the objects are connected via offsets.