TODO
sandeep-krishnamurthy opened this issue · comments
- Pre-trained weights for networks - It would be helpful if pre-trained weights are provided as an option to load it into the networks rather than just asking readers to always train from scratch.
- Hybridizable Blocks - All implementations use Sequential block i.e., non-hybridizable. May it would be useful to convert all implementation to hybridizable and allow exporting the symbol, params.