Skip to content

Aleph-Alpha-Research/hat-splitter

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

62 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

hat-splitter

This is the home of the Hierarchical Autoregressive Transformer (HAT) splitting rule. You can use this to implement training and inference of HAT models.

We expose it as a Rust crate with Python bindings so that (1) we can use the same splitting rule in both languages; and (2) we improve performance for inference.

Performance

The following pytest benchmark result demonstrates the performance advantage of this Rust implementation over the previous pure Python implementation.

-------------------------------------------------------
Name (time in ms)                         Mean         
-------------------------------------------------------
test_benchmark_hat_splitter            41.3459 (1.0)   
test_benchmark_scaling_splitter     2,415.7853 (58.43) 
-------------------------------------------------------

Development

See the Python bindings README.md for development instructions for the Python bindings.

Release process

  1. Update the version in Cargo.toml. Commit and push to main.
  2. Tag the commit with the new version, e.g., git tag v0.1.0.
  3. Push the tag to the remote. CI will take care of the rest.

License

See LICENSE.

About

Rust crate with Python bindings for the HAT splitting rule

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Contributors 2

  •  
  •