RouteOptions Struct
Options for Model::graph() — controls how the model assembles into a Graph. More...
Declaration
Included Headers
Public Member Attributes Index
| bool | include_input = false |
|
Include a public nodes::Input Node at the head. More... | |
| bool | include_output = false |
|
Include a public nodes::Output Node at the tail. More... | |
| bool | expose_all_outputs = false |
|
Expose individual model output endpoints when the model route has multiple physical outputs. More... | |
| std::string | upstream_name |
|
Element name of the upstream Node feeding the model (default: "decoder"). More... | |
| std::string | name_suffix |
|
Suffix appended to generated element names. More... | |
| std::string | buffer_name |
|
Optional buffer pool name (used by advanced memory configurations). More... | |
| VerboseOptions | verbose |
| std::string | processcvu_requested_run_target = "AUTO" |
|
Requested backend for model-managed processcvu generic-EV stages. More... | |
| ProcessCvuOptions | processcvu |
|
Simple placement for model-managed processcvu stages when this Graph/Runner is built. More... | |
| ProcessMlaOptions | processmla |
|
MLA stage execution options. More... | |
| PreparedRunnerOptions | prepared_runner |
|
Experimental prepared-route runner. More... | |
| int | async_queue_depth = 0 |
|
Depth for internally inserted async queue2 elements. More... | |
Description
Options for Model::graph() — controls how the model assembles into a Graph.
By default model.graph() returns a reusable model fragment with open named endpoints. Set include_input and/or include_output when you intentionally want the returned Graph to contain public nodes::Input / nodes::Output boundary nodes.
Definition at line 282 of file Model.h.
Public Member Attributes
async_queue_depth
|
Depth for internally inserted async queue2 elements.
0 inherits from Model::Options/framework default.
Definition at line 327 of file Model.h.
buffer_name
|
Optional buffer pool name (used by advanced memory configurations).
expose_all_outputs
|
Expose individual model output endpoints when the model route has multiple physical outputs.
The default public Graph contract is intentionally ML-friendly: a model produces one aggregate output Sample, normally a TensorSet containing all model tensors. That remains true for optimized routes where several logical tensors are packed into one physical buffer.
Set this to true only when advanced/debug code needs to address truly separate physical model outputs by name. If the route has a single physical output buffer, this flag is ignored and the Graph still exposes one aggregate output endpoint.
Definition at line 297 of file Model.h.
include_input
|
Include a public nodes::Input Node at the head.
Definition at line 283 of file Model.h.
include_output
|
Include a public nodes::Output Node at the tail.
Definition at line 284 of file Model.h.
name_suffix
|
Suffix appended to generated element names.
Definition at line 300 of file Model.h.
prepared_runner
|
Experimental prepared-route runner.
Non-default values take priority over Model::Options.
processcvu
|
Simple placement for model-managed processcvu stages when this Graph/Runner is built.
Non-AUTO values take priority over Model::Options and the legacy coarse target above.
processcvu_requested_run_target
|
Requested backend for model-managed processcvu generic-EV stages.
"AUTO" (default) resolves in core during manifest rendering and prefers A65 when the stage supports it; otherwise EV74. Set to "EV74" or "A65" to force a specific backend.
Definition at line 310 of file Model.h.
processmla
|
MLA stage execution options.
Non-default values take priority over Model::Options.
upstream_name
|
Element name of the upstream Node feeding the model (default: "decoder").
Definition at line 298 of file Model.h.
verbose
The documentation for this struct was generated from the following file:
Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.