Commits on Apr 19, 2023

1. Add jagged slice op for cpu (pytorch#1690) …

   Summary:
   Pull Request resolved: pytorch#1690
   
   The context why this is needed is as follows
   1) For really long sparse features we want to split them into multiple chunks that can be fed into the model
   2) Slicing requires users to require per row start point & a maximum L.
   
   Based on these requirements, a custom op mimicing the slice semantics of a normal tensor works best.
   
   An example usage using pseudo code
   
   ```
   input_jagged_tensor = [[1, 2, 3, 4], [1, 2, 3], [1, 2, 3, 4, 5, 6], [1], [1, 2]]
   start = [0, 0, 0, 0, 0]
   slice_length = 3
   
   >> jagged_slice(input_jagged_tensor, start, slice_length)
   
   output_jagged_tensor = [[1, 2, 3], [1, 2, 3], [1, 2, 3], [1], [1, 2]]
   
   ```
   
   A corresponding operation for dense tensor would look like
   ```
   dense_tensor = torch.randn((8, 10))
   slice_dense_tensor = dense_tensor[:, 1:3]
   ```
   
   Reviewed By: sryap
   
   Differential Revision: D44299744
   
   fbshipit-source-id: 982ba26987643cbb778fdfac7d9ac4fd79d8b369

   

   Devashish Tyagi authored and facebook-github-bot committed Apr 19, 2023
   Configuration menu

   • View commit details
   • Copy full SHA for 03c6990
   • Browse repository at this point

   Copy the full SHA

   03c6990 View commit details

   Browse the repository at this point in the history