Open wvabrinskas opened 5 months ago
Looked into using InputStream
with JSONSerialization
however it's still allocating a large chunk of memory. Will try breaking the allocations by Layer so the JSON isnt all in memory at once.
So it turns out I was encoding the JSON as swift Data object TWICE. causing excess memory allocations.
Maybe use some Quantization?
Create a new compressed form of the smodel format for model exports. Model sizes can quickly get really large. Having a compressed version will allow for larger models to be expressed smaller on disk.