Compute and memory based artificial intelligence model partitioning using intermediate representation

A semiconductor apparatus comprising:

• one or more substrates;
• and
• logic coupled to the one or more substrates, wherein the logic is implemented in one or more of configurable logic or fixed-functionality logic hardware, the logic coupled to the one or more substrates to:
• convert an artificial intelligence (AI) model graph into an intermediate representation, wherein the intermediate representation is hardware independent;
• determine computations associated with the AI model graph based on a plurality of layers identified from the intermediate representation, wherein a layer is a node of the AI model graph;
• partition the intermediate representation of the AI model graph into a plurality of subgraphs based on the computations, the subgraphs being associated with the subgraph memory resources and a plurality of hardware devices;
• determine whether to readjust the plurality of subgraphs based on the subgraph memory resources associated with the plurality of subgraphs and memory capacities of the plurality of hardware devices;
• identify a total compute workload of the AI model graph based on the computations associated with the AI model graph;
• identify a compute value based on the total compute workload of the AI model graph divided by a total number of the plurality of subgraphs;
• and
• iteratively add layers to each respective subgraph of the plurality of subgraphs while a total compute workload of the respective subgraph is less than the compute value.

At least one non-transitory computer readable storage medium comprising a set of executable program instructions, which when executed by a computing system, cause the computing system to:

• convert an artificial intelligence (AI) model graph into an intermediate representation, wherein the intermediate representation is hardware independent;
• determine computations associated with the AI model graph based on a plurality of layers identified from the intermediate representation, wherein a layer is a node of the AI model graph;
• partition the intermediate representation of the AI model graph into a plurality of subgraphs based on the computations, the subgraphs being associated with the subgraph memory resources and a plurality of hardware devices;
• determine whether to readjust the plurality of subgraphs based on the subgraph memory resources associated with the plurality of subgraphs and memory capacities of the plurality of hardware devices;
• identify a total compute workload of the AI model graph based on the computations associated with the AI model graph;
• identify a compute value based on the total compute workload of the AI model graph divided by a total number of the plurality of subgraphs;
• and
• iteratively add layers to each respective subgraph of the plurality of subgraphs while a total compute workload of the respective subgraph is less than the compute value.

A method comprising:

• converting an artificial intelligence (AI) model graph into an intermediate representation, wherein the intermediate representation is hardware independent;
• determining computations associated with the AI model graph based on a plurality of layers identified from the intermediate representation, wherein a layer is a node of the AI model graph;
• partitioning the intermediate representation of the AI model graph into a plurality of subgraphs based on the computations, the subgraphs being associated with the subgraph memory resources and a plurality of hardware devices;
• determining whether to readjust the plurality of subgraphs based on the subgraph memory resources associated with the plurality of subgraphs and memory capacities of the plurality of hardware devices;
• identifying a total compute workload of the AI model graph based on the computations associated with the AI model graph;
• identifying a compute value based on the total compute workload of the AI model graph divided by a total number of the plurality of subgraphs;
• and
• iteratively adding layers to each respective subgraph of the plurality of subgraphs while a total compute workload of the respective subgraph is less than the compute value.