Executing queries in computing systems using execution plans generated by generative artificial intelligence models
Certain aspects provide techniques and apparatus for executing queries in a computing system using machine learning models. An example method generally includes receiving a plan to satisfy a request in the computing system and event log data associated with execution of the plan. The plan generally specifies a first plurality of actions to be performed by the computing system at a first level of granularity. Using a plan refinement machine learning model, a refined plan is generated when the event log data indicates that execution of the generated plan results in one or more execution errors and the one or more execution errors are solvable. Generally, the refined plan specifies a second plurality of actions to be performed by the computing system at a second level of granularity, the second level of granularity being finer than the first level of granularity.
1 . A processing system for invoking functions in a computing system using machine learning models, comprising:
at least one memory having executable instructions stored thereon; and
one or more processors configured to execute the executable instructions to cause the processing system to:
receive a plan to satisfy a request in the computing system and event log data associated with execution of the plan, the plan specifying a first plurality of actions to be performed by the computing system at a first level of granularity;
generate, using a plan refinement machine learning model, a refined plan when the event log data indicates that execution of the generated plan results in one or more execution errors and the one or more execution errors are solvable, the refined plan specifying a second plurality of actions to be performed by the computing system at a second level of granularity, the second level of granularity being finer than the first level of granularity; and
determine that the one or more execution errors are solvable, wherein to determine that the one or more execution errors are solvable, the one or more processors are configured to cause the processing system to determine, based on a generative artificial intelligence model and a set of prompt-engineered instructions, that a complexity associated with resolving the one or more execution errors is less than a threshold complexity level.
2 . The processing system of claim 1 , wherein the plan refinement machine learning model comprises a generative model trained based on reinforcement learning with assistant feedback and wherein the assistant feedback comprises an indication of whether the refined plan generated by the plan refinement machine learning model is executable by an assistant configured to execute on another computing system.
3 . The processing system of claim 2 , wherein the generative model is further trained to minimize a length of the refined plan.
4 . The processing system of claim 1 , wherein to generate the refined plan, the one or more processors are configured to cause the processing system to iteratively generate refined plans until one of the generated refined plans is accepted as an executable plan.
5 . The processing system of claim 1 , wherein to generate the refined plan, the one or more processors are configured to cause the processing system to iteratively generate refined plans until a threshold number of plan refinement iterations have been executed by the plan refinement machine learning model.
6 . The processing system of claim 1 , wherein the plan refinement machine learning model comprises a tree model trained based on generating a tree data structure including a plurality of nodes, wherein each node in the tree data structure comprises one of a plan or one or more refined plans, and wherein the tree data structure has a maximum depth equal to a threshold number of plan refinement iterations.
7 . A processor-implemented method for invoking functions in a computing system using machine learning models, comprising:
receiving a plan to satisfy a request in the computing system and event log data associated with execution of the plan, the plan specifying a first plurality of actions to be performed by the computing system at a first level of granularity;
generating, using a plan refinement machine learning model, a refined plan when the event log data indicates that execution of the generated plan results in one or more execution errors and the one or more execution errors are solvable, the refined plan specifying a second plurality of actions to be performed by the computing system at a second level of granularity, the second level of granularity being finer than the first level of granularity;
determining that the one or more execution errors are solvable, wherein the determining comprises determining, based on a generative artificial intelligence model and a set of prompt-engineered instructions, that a complexity associated with resolving the one or more execution errors is less than a threshold complexity level.
8 . The method of claim 7 , wherein the plan refinement machine learning model comprises a generative model trained based on reinforcement learning with assistant feedback and wherein the assistant feedback comprises an indication of whether the refined plan generated by the plan refinement machine learning model is executable by an assistant configured to execute on another computing system.
9 . The method of claim 8 , wherein the generative model is further trained to minimize a length of the refined plan.
10 . The method of claim 7 , wherein generating the refined plan comprises iteratively generating refined plans until one of the generated refined plans is accepted as an executable plan.
11 . The method of claim 7 , wherein generating the refined plan comprises iteratively generating refined plans until a threshold number of plan refinement iterations have been executed by the plan refinement machine learning model.
12 . The method of claim 7 , wherein the plan refinement machine learning model comprises a tree model trained based on generating a tree data structure including a plurality of nodes, wherein each node in the tree data structure comprises one of a plan or one or more refined plans, and wherein the tree data structure has a maximum depth equal to a threshold number of plan refinement iterations.
13 . A non-transitory computer-readable medium having executable instructions stored thereon which, when executed by one or more processors, performs an operation for invoking functions in a computing system using machine learning models, the operation comprising:
receiving a plan to satisfy a request in the computing system and event log data associated with execution of the plan, the plan specifying a first plurality of actions to be performed by the computing system at a first level of granularity;
generating, using a plan refinement machine learning model, a refined plan when the event log data indicates that execution of the generated plan results in one or more execution errors and the one or more execution errors are solvable, the refined plan specifying a second plurality of actions to be performed by the computing system at a second level of granularity, the second level of granularity being finer than the first level of granularity; and
determining that the one or more execution errors are solvable, wherein the determining comprises determining, based on a generative artificial intelligence model and a set of prompt-engineered instructions, that a complexity associated with resolving the one or more execution errors is less than a threshold complexity level.
14 . The computer-readable medium of claim 13 , wherein the plan refinement machine learning model comprises a generative model trained based on reinforcement learning with assistant feedback and wherein the assistant feedback comprises an indication of whether the refined plan generated by the plan refinement machine learning model is executable by an assistant configured to execute on another computing system.
15 . The computer-readable medium of claim 14 , wherein the generative model is further trained to minimize a length of the refined plan.
16 . The computer-readable medium of claim 13 , wherein generating the refined plan comprises iteratively generating refined plans until one of the generated refined plans is accepted as an executable plan.
17 . The computer-readable medium of claim 13 , wherein generating the refined plan comprises iteratively generating refined plans until a threshold number of plan refinement iterations have been executed by the plan refinement machine learning model.