Maml update: synthesize data and parameterize (#465)
Summary:
Generate model values instead of loading from 10mb file. Also model sizes are now partially parameterized.
This is a nop change wrt performance but simplifies future model investigation.
We can remove batch.pt from repo once we're happy with with model performance.
Pull Request resolved: https://github.com/pytorch/benchmark/pull/465
Reviewed By: H-Huang
Differential Revision: D31152978
Pulled By: xuzhao9
fbshipit-source-id: 0b28dd305b5b921a03a4031ae6507ce6f26025e9
