MagmaDNN / magmadnn

MagmaDNN: a simple deep learning framework in c++
MIT License
45 stars 7 forks source link

[FEATURE REQUEST] Easier integration of model parallelism #24

Open Dando18 opened 5 years ago

Dando18 commented 5 years ago

Is your feature request related to a problem? Please describe. Models are not distributed in parallel by default (i.e. when they exceed device memory). Thus the user must first encounter an Out of Memory error and then employ some model parallelism techniques.

Describe the solution you'd like The Model class or Compute Graph should be able to detect this and distribute accordingly.