Genienlp

=======

* Add support for end-to-end dialogue training and evaluation [251]
* Add dialogue loop enabling interactions with dialogue models over the command-line [276]
* Adding hybrid alignment using a two-step translation approach [263]
* Adding a new batching algorithm [290]
* Adding new metrics for dialogues [270]
* Major code overhauling and cleanups [252]
* Add script to compute evaluation metrics on a file generated from genienlp predict [253]
* Add direct prediction for HuggingFace models [287]
* Fix model saving code to avoid double memory consumption [219]
* Updating bootleg analysis scripts [235]
* Adding contribution guidelines and updating documentation [289, 290]
* Various bug fixes [229, 234,242, 250, 259, 285]

=======

* Added adafactor optimizer [204]
* Added option to compute and log validation loss [204]
* Multiple changes to improve alignment in translation, better heuristics for dates and numbers [204]
* Added support for changing generation arguments in server requests [204]
* Misc. code upgrades and bug fixes [211, 216, 217]
* Updated dependencies (major update for sacrebleu) [202, 203, 205, 206, 209, 210, 212, 214, 218].

=======

* Fix installation from pip [201].

=======

* Added support for sequence classification tasks [176].
* NED was refactored and cleaned up. Bootleg can now accept an optional file to normalize
types. Support for ElasticSearch in the naive NED was removed. Option names were simplified
and documentation was improved [173, 183, 188, 191].
* kfserving will now terminate with an error in case of CUDA error, allowing graceful
restart [190].

=====

* Added support for token classification tasks, and added AmbigQA and CrossNER tasks [128].
* Added support for beam search in server mode [141].
* Misc bug fixes [142, 143].
* Updated dependencies [133, 134, 136, 138, 140, 144, 145].
* Build system and code style fixes [146].

=======

* Translation code is now migrated to the main genienlp codebase under `almond_translate` task [98, 105].
* Changed data batching code to account for sequence output lengths too [130]
* Pipenv is removed [104].
* Bumped Bootleg version to 1.0.1 [124].
* Bumped Transformers version to 4.5.1 [114, 127].
* Misc code upgrades and bug fixes [115, 102, 123].