Problem
- Targets may require adjustments for the
mlops: parameters. This can be solved by using $variables$ in the mlops: section.
- Some MLOps system may require additional information about the model. This can be solved by a flexible
key: value list under mlops.model:. Note that clayer is really just a key/value pair - there is extra handling required for clayer.
Proposed changes
The mlops: node could then look like below:
mlops:
description: TinyCNN int8 image classifier for Ethos-U85
npu:
type: $NPU$
macs: $Ethos-Macs$
vela:
system: $Ethos_system$ # Ethos_U85_SYS_DRAM_Mid # system-config from the Vela config
memory: Shared_Sram # memory-mode from the Vela config
model:
clayer: $AI-Layer$
key1: value1
key2: $variable2$
simulator:
target: SSE-320-U85
In case that a variable is undefined, it should expand into an empty string (meaning that the mlops key has no value in the *.cbuild-mlops.yml file.
Problem
mlops:parameters. This can be solved by using$variables$in themlops:section.key: valuelist undermlops.model:. Note that clayer is really just a key/value pair - there is extra handling required for clayer.Proposed changes
The
mlops:node could then look like below:In case that a variable is undefined, it should expand into an empty string (meaning that the mlops key has no value in the
*.cbuild-mlops.ymlfile.