nntrainer
nntrainer copied to clipboard
context.request* should gaurantee index
https://github.com/nnstreamer/nntrainer/blob/f26937ee48243bb099edd37538d90054cd735d3d/nntrainer/layers/layer_context.h#L144-L145
above todo is critical when guaranteeing serialized weight layout, without those kind of deterministic assumption, we cannot load weight outside context of layer.
eg) fully connected layer has weight - maps to a, bias - maps to b When we load the weight from different place, the layer weights will maps to weight - a', bias - b'. we have to guarantee a == a', b == b'. If this indexes are determined per run, we cannot never make such assumption.
The implementation implemented in a way it is deterministic. We just have to make it as the api behavior :)
:octocat: cibot: Thank you for posting issue #1814. The person in charge will reply soon.