ARTICLE AD BOX
So I am building an AI model and I was trying to implement one optimization in it so it gets faster.
The AI model is for DAG of code. The high level idea of the optimization is that in a DAG if picks out common nodes ,for example: varA->funcA->funcB->funcC->varB and varZ->funcA->funcB->funcC->varY, what I want to do is take these common nodes and make it a single representation to improve memory and speed, something like varA->funcX->funcB and varZ->funcX->varY and define funcX=funcA->funcB->funcC after converting these things into numbers, and then convert this funcX and these conversions to a representation using embeddings but there is a very likely chance that the model would have never seen funcX so I want the representation to be like so that it tells the model, hey I am just a optimization trick my info is present at the last of this sample without increasing the memory or replacing the value or funcX with funcA->funcB->funcC. Is that possible?
