Commit Graph

41 Commits

Author SHA1 Message Date
mehrad 701f4d003c Merge branch 'master' into wip/mehrad/master 2021-01-05 00:19:41 -08:00
Sina 15dfff170e Update `tensorboardX` version
Fixes a bug with `purge_step`
2021-01-04 20:10:07 -08:00
mehrad 46b4ddf471 Merge branch 'master' into wip/mehrad/master 2020-12-30 11:14:32 -08:00
Sina 547749e4f9 Add support for Loss Truncation
Loss Truncation is introduced in https://arxiv.org/abs/2004.14589.
It can potentially improve the quality of our natural language generation
tasks like paraphrasing and `almond_dialogue_nlg`.
2020-12-24 22:30:35 -08:00
mehrad 0b28e52d7d Merge branch 'wip/mehrad/v0.5.0' into wip/mehrad/master 2020-12-23 12:30:23 -08:00
mehrad 6587127c56 Update pipfiles 2020-12-22 12:00:21 -08:00
Sina b1473d769c Change pytorch version to 1.7.1
This version of Pytorch fixed a bug that caused severe memory issues
when `transformers` was used with Pytorch's mixed precision
2020-12-20 00:08:03 -08:00
Sina 89b680e4c7 Change transformers version to 4.1.1
This version has the implementation for
diverse beam search (https://arxiv.org/abs/1610.02424).
2020-12-19 22:59:21 -08:00
Giovanni Campagna 259387848e Restore sentencepiece dependency
We actually need it for XLM-R
2020-12-18 09:23:04 -08:00
Giovanni Campagna db8e9fb1ee Merge remote-tracking branch 'origin/master' into wip/thingtalk2 2020-12-14 21:46:51 -08:00
mehrad 2be50ff8ce update Pipfiles 2020-12-10 11:54:45 -08:00
Giovanni Campagna cd6cf37a9e Remove revtok 2020-12-05 17:37:39 -08:00
mehrad 2ac537d1e4 Update pipfiles to use python 3.8 2020-12-03 17:25:00 -08:00
mehrad 8b644cca02 bump transformers version to 3.5.1 2020-12-03 12:50:34 -08:00
s-jse 37f80df956
Update minimum pytorch version to 1.6 (#44)
We have experienced some issues with the transformers library
when pytorch <1.6 is installed.
2020-11-03 14:24:39 -08:00
mehrad 5139c71655 Update python-dateutil version 2020-08-10 15:51:41 -07:00
mehrad 91a7ea0e2e Pipfile: set numpy version 2020-07-16 13:41:25 -07:00
mehrad 140a5e0acd Pipfile: add matplotlib and seaborn 2020-07-16 13:41:25 -07:00
mehrad 40dfeadd2b upgrade transformers to 2.11 2020-07-16 13:38:48 -07:00
Sina 1ed4c93a62 Update transformers to version 2.9
This resolves issues transformers had with BART until version 2.8
2020-05-11 22:15:55 -07:00
Giovanni Campagna bb6018ba01 Add Pipfile script
So you can run "pipenv run genienlp" instead of "pipenv run python3 -m genienlp"
2020-03-19 10:37:16 -07:00
Giovanni Campagna f971c31dde Merge remote-tracking branch 'origin/master' into wip/contextual 2020-03-16 12:42:21 -07:00
Sina 190f953833 newer versions of transformers package are not backward compatible 2020-03-02 16:47:57 -08:00
mehrad c1487ce1db minor changes 2020-02-20 11:36:00 -08:00
mehrad 8145952a25 XLM-R model as encoder 2020-02-19 15:04:50 -08:00
Giovanni Campagna 4de1b58e4c Switch back to python 3.6
We don't have 3.7 in the docker images
2020-02-13 09:22:48 -08:00
Giovanni Campagna 448cd65123 pipenv: install scikit-learn
To avoid a warning at runtime
2020-02-10 17:10:53 -08:00
Giovanni Campagna c89e069f55 Make radam optional
We cannot depend on radam in the pypi package
2020-01-29 10:15:25 -08:00
Giovanni Campagna 91aff28abc Add support for radam optimizer 2020-01-19 19:40:01 -08:00
Giovanni Campagna 47479a3b65 Remove ujson library
Just use the standard library, the extra dependency is not worth it
2020-01-16 21:01:31 -08:00
Giovanni Campagna 3da2e8f805 Remove dateutil library
So much just get the current time in ISO format...
2020-01-16 20:59:17 -08:00
Giovanni Campagna aa6d0f0c3f Update dependencies
Add version specs to everything, and add transformers
2020-01-16 20:54:24 -08:00
Giovanni Campagna bbcdb1a3e2 Add a BERT-based numericalizer
Compatible with wordpiece tokenization used by BERT, and also
with Almond's unique requirements around how tokens are split
2020-01-15 22:27:58 -08:00
Giovanni Campagna c0b4d582b7 Add missing dependency to Pipfile 2020-01-14 12:58:02 -08:00
Giovanni Campagna d04fe60e28 Update to Python 3.7
Python 3.6 is obsolete
2020-01-13 17:15:30 -08:00
Giovanni Campagna 27b8dca79f Remove a whole lot of unused Almond code
Remove the "thingpedia_as_context" option, and the different ThingTalk
grammars. They were unused, and also unlikely to be helpful, after
several experiments.
2019-12-10 18:03:01 -08:00
Giovanni Campagna 4cf9a1941f Remove ELMo embeddings
Like Cove, ELMo embeddings are also obsolete, and the removal
allows us to remove the large allennlp library.
2019-12-10 16:38:23 -08:00
Giovanni Campagna e7481840f7 Remove cove
Cove is an obsolete form of pretraining, of dubious utility. It
depends on an obsolete, unmaintained, library, and it causes
hacks in the code. Clean up.
2019-12-10 16:34:35 -08:00
mehrad 7555ec6b82 master updates + additional tweaks 2019-03-21 11:20:48 -07:00
mehrad 9e08d78535 updating outdated Pipfiles 2019-03-12 16:44:23 -07:00
mehrad 19c9a100d5 adding Pipfiles 2019-03-12 13:30:27 -07:00