InferenceTerminalPolicy Struct
Where the inference pipeline should terminate. More...
Declaration
Included Headers
Public Member Attributes Index
| bool | mla_only = false |
|
If true, terminate at the MLA stage; skip postprocess entirely. More... | |
| std::optional< std::size_t > | last_stage_index |
|
Stop after this stage index (0-based, inclusive). More... | |
| std::optional< std::string > | last_stage_name |
|
Stop after the named stage (matches MPK manifest stage name). More... | |
| std::optional< std::string > | last_plugin_id |
|
Stop after the first stage using this plugin ID. More... | |
| std::optional< std::string > | last_processor |
|
Stop after the first stage running on this processor (CVU/MLA/APU). More... | |
Description
Where the inference pipeline should terminate.
Most pipelines run preprocess → MLA → postprocess. This policy lets advanced users stop earlier — for example, to inspect the raw MLA output before postprocessing, or to chain two models where the first one only needs to run through the MLA stage.
Definition at line 155 of file Model.h.
Public Member Attributes
last_plugin_id
|
Stop after the first stage using this plugin ID.
Definition at line 161 of file Model.h.
last_processor
|
Stop after the first stage running on this processor (CVU/MLA/APU).
last_stage_index
|
Stop after this stage index (0-based, inclusive).
last_stage_name
|
Stop after the named stage (matches MPK manifest stage name).
mla_only
The documentation for this struct was generated from the following file:
Generated via doxygen2docusaurus 2.0.0 by Doxygen 1.9.8.