v2.2.0

What's new

Added 🎉

• Evaluating RC task card and associated LERC model card (@anthonywchen).
• Compatibility with PyTorch 1.8.
• Dataset reader for SuperGLUE BoolQ.

Changed ⚠️

• Add option combine_input_fields in SnliDatasetReader to support only having "non-entailment" and "entailment" as output labels (@leo-liuzy).
• Made all the models run on AllenNLP 2.1.
• Add option ignore_loss_on_o_tags in CrfTagger to set the flag outside its forward function (@tarohi24).

Fixed ✅

• Fixed allenai/allennlp#4745 (@leo-liuzy).
• Updated QaNet and NumericallyAugmentedQaNet models to remove bias for layers that are followed by normalization layers.
• Updated the model cards for rc-naqanet, vqa-vilbert and ve-vilbert.
• Predictors now work for the vilbert-multitask model.

Commits

f5c157b Prepare for release v2.2.0
9a237c3 fixes for callback refactor (#233)
8aabfe5 add BoolQ dataset reader (#228)
ab2a0d1 Added model card for LERC and task card for evaluating-RC (#226)
986265d Add ignore_loss_on_o_tags to CrfTagger.__init__() (#230)
20bc25a Demo Sweep 2 (#227)
2f798c0 Fix sanity checks for models (#224)
3624fdc Update torch requirement from <1.8.0,>=1.7.0 to >=1.7.0,<1.9.0 (#225)
ec115c6 Fix NextTokenLM bug (#213)
ff68203 Add option to collapse non-entailment labels in SNLI dataset reader (#214)
743e0ee Fix QaNet and NaQaNet (#223)
769cda0 small doc fix in CopyNet
ef51b67 Fixes Multitask predictions for the vision models (#218)

v1.5.0

What's new

Added 🎉

• BART model now adds a predicted_text field in make_output_human_readable that has the cleaned text corresponding to predicted_tokens.

Fixed ✅

• Made label parameter in TransformerMCReader.text_to_instance optional with default of None.
• Fixed OpenIePredictor.predict_json so it treats auxiliary verbs as verbs
  when the language is English.

Commits

6355102 small doc fix in CopyNet
14ffde8 Fix prediction for models using TransformerMCReader (#222)
c411e37 copy over TAG_MAP from spaCy (#219)
8e700bb Add predicted_text field to BART model output (#217)
845ab1e OpenIE fix (#206)
48150a3 Updating modelcards for ModelUsage (#209)

v2.1.0

What's new

Changed ⚠️

• coding_scheme parameter is now deprecated in Conll2000DatasetReader, please use convert_to_coding_scheme instead.

Added 🎉

• BART model now adds a predicted_text field in make_output_human_readable that has the cleaned text corresponding to predicted_tokens.

Fixed ✅

• Made label parameter in TransformerMCReader.text_to_instance optional with default of None.
• Updated many of the models for version 2.1.0. Fixed and re-trained many of the models.

Commits

1c006c5 Demo model sweep (#216)
1d44f1a Fix prediction for models using TransformerMCReader (#222)
9128347 Update mkdocs-material requirement from <6.3.0,>=5.5.0 to >=5.5.0,<7.1.0 (#221)
1b32cae Bump mypy from 0.800 to 0.812 (#220)
c645f78 copy over TAG_MAP from spaCy (#219)
431e5d1 Bump conllu from 4.3.1 to 4.4 (#215)
0634412 Add predicted_text field to BART model output (#217)
e51c051 Bump conllu from 4.3 to 4.3.1 (#212)
c258a83 Introduce convert_to_coding_scheme and make coding_scheme deprecated in CoNLL2000DatasetReader (#211)
dfded46 Remove conll2003_test.py from allennlp-models (#210)
95eb53e Updating modelcards for ModelUsage (#209)

v2.0.1

What's new

Fixed ✅

• Fixed OpenIePredictor.predict_json so it treats auxiliary verbs as verbs
  when the language is English.

Commits

ca7184c tick version for nightly release
da80d8f adding more examples for lm (#208)
0413882 full image paths (#207)
35cfb57 OpenIE fix (#206)
01a4b28 adding vqa model card, renaming ve modelcard (#205)
42c0450 Description and citation as markdown (#190)
8fd7aaa fix ALLENNLP_COMMIT_SHA

v1.4.1

What's new

Note: This release is mainly for the AllenNLP demo.

Changed ⚠️

• Updated various model cards and task cards.

Commits

80a985b adding more examples for lm (#208)
90d2418 full image paths (#207)
f498854 adding vqa model card, renaming ve modelcard (#205)
0413a7e Description and citation as markdown (#190)

v2.0.0

For the main differences between AllenNLP 2.0 and AllenNLP 1.x, check out the release announcement at https://github.com/allenai/allennlp/releases/tag/v2.0.0!

What's new (since v2.0.0rc1)

Fixed ✅

• Made the training configs compatible with the tensorboard logging changes in the main repo

Commits

f0aff04 fixes for tensorboard changes (#204)
60965a9 ci quick fix
167728c ci quick fix
2d4c437 ci quick fix
cbbd58d tick version for nightly release
bde73ce changing text->sentence, more examples (#203)
a507f5d Merging vision into main (#171)
d1a54d7 adding task card for language modeling (#201)
75e9c1a Bump mypy from 0.790 to 0.800 (#200)
392cee7 fix mkdocs config (#198)
5ddc8aa Fix BART (#195)
9472904 Fix TransformerQA distributed dead-lock bug (#193)
54ccc82 Bump conllu from 4.2.2 to 4.3 (#194)
8f18a17 Update mkdocs-material requirement from <6.2.0,>=5.5.0 to >=5.5.0,<6.3.0 (#185)
2644f6a Fix small bugs in RC model (#191)
92768ae Moving modelcard and taskcard abstractions to main repo (#186)
92f1eea fix requirements.txt
641c8ef fix navbar link
70cd713 Fixes for master->main rename (#188)
c62a347 ensure latest allennlp commit always used (#187)
b8c858a Bump conllu from 4.2.1 to 4.2.2 (#180)
4b16dbc fix cache volume (#184)

v1.4.0

What's new

Removed 👋

• Moving ModelCard and TaskCard abstractions out of the models repository.

Changed ⚠️

• 'master' branch renamed to 'main'
• SquadEmAndF1 metric can now also accept a batch of predictions and corresponding answers (instead of a single one)
  in the form of list (for each).

Fixed ✅

• Fixed the potential for a dead-lock when training the TransformerQA model on multiple GPUs
  when nodes receive different sized batches.

Fixed ✅

• Fixed BART. This implementation had some major bugs in it that caused poor performance during prediction.

Commits

1c300d8 Prepare for release v1.3.1
0402f6d changing text->sentence, more examples (#203)
7652c6c Make the 1.x branch work in CI
d1a54d7 adding task card for language modeling (#201)
75e9c1a Bump mypy from 0.790 to 0.800 (#200)
392cee7 fix mkdocs config (#198)
5ddc8aa Fix BART (#195)
9472904 Fix TransformerQA distributed dead-lock bug (#193)
54ccc82 Bump conllu from 4.2.2 to 4.3 (#194)
8f18a17 Update mkdocs-material requirement from <6.2.0,>=5.5.0 to >=5.5.0,<6.3.0 (#185)
2644f6a Fix small bugs in RC model (#191)
92768ae Moving modelcard and taskcard abstractions to main repo (#186)
92f1eea fix requirements.txt
641c8ef fix navbar link
70cd713 Fixes for master->main rename (#188)
c62a347 ensure latest allennlp commit always used (#187)
b8c858a Bump conllu from 4.2.1 to 4.2.2 (#180)
4b16dbc fix cache volume (#184)

v2.0.0rc1

This is the first (and hopefully only) release candidate for AllenNLP 2.0. Please note that this is a release candidate, and the APIs are still subject to change until the final 2.0 release. The big addition is the first item, models for several tasks that combine language and vision. Hopefully these will serve as an example for many future models.

What's new

Added 🎉

• Dataset readers, models, metrics, and training configs for VQAv2, GQA, and Visual Entailment

Fixed ✅

• Fixed training_configs/pair_classification/bimpm.jsonnet and training_configs/rc/dialog_qa.jsonnet
  to work with new data loading API.
• Fixed the potential for a dead-lock when training the TransformerQA model on multiple GPUs
  when nodes receive different sized batches.
• Fixed BART. This implementation had some major bugs in it that caused poor performance during prediction.

Removed 👋

• Moving ModelCard and TaskCard abstractions out of the models repository.

Changed ⚠️

• master branch renamed to main
• SquadEmAndF1 metric can now also accept a batch of predictions and corresponding answers (instead of a single one)
  in the form of list (for each).

Commits

07fc94e Moving vision models (#199)
493e961 Merge remote-tracking branch 'origin/main' into vision
392cee7 fix mkdocs config (#198)
5ddc8aa Fix BART (#195)
9472904 Fix TransformerQA distributed dead-lock bug (#193)
64fdd59 Merge branch 'main' into vision
953822d fix piqa reader test
54ccc82 Bump conllu from 4.2.2 to 4.3 (#194)
5c8ec4f merge conflicts
8f18a17 Update mkdocs-material requirement from <6.2.0,>=5.5.0 to >=5.5.0,<6.3.0 (#185)
2644f6a Fix small bugs in RC model (#191)
d2d7b55 fixes for data loading updates in core (#192)
debf15a VE model card (#189)
92768ae Moving modelcard and taskcard abstractions to main repo (#186)
92f1eea fix requirements.txt
641c8ef fix navbar link
70cd713 Fixes for master->main rename (#188)
c62a347 ensure latest allennlp commit always used (#187)
e80f702 Merge branch 'master' into vision
b8c858a Bump conllu from 4.2.1 to 4.2.2 (#180)
4b16dbc fix cache volume (#184)
ed564d2 Merge remote-tracking branch 'origin/master' into vision
d1d37de fix SNLI reader
f5ba4a6 fix LM training config
47401fa fix test-install CI job (#172)
af4a7e2 merge with master
058815e update with master
8b908e0 fix merge conflicts
5e778ac Training config test fix (#157)
bd47e3b Merge branch 'master' into vision
8385f47 fix TransformerQA predictor test (#145) (#146)
1c7749d fix merge conflicts
ac4fc45 Merge remote-tracking branch 'origin/master' into vision
09fe0b0 improve generation dataset readers (#122)
48bceca updates
455ac1f ensure vision CI runs on every commit (#120)
4e48a16 Updates for new data loading API (#101)
10b9db3 Merge branch 'master' into vision
182a7bf Merge branch 'master' into vision
e56b22a run on vision PRs
c1178c3 temporarily patch allennlp to data-loading branch

v1.3.0

What's new

Fixed ✅

• Fix an index bug in BART prediction.
• Add None check in PrecoReader's text_to_instance() method.
• Fixed SemanticRoleLabelerPredictor.tokens_to_instances so it treats auxiliary verbs as verbs
  when the language is English
• Added missing folder for taskcards in setup.py

Added 🎉

• Added link to source code to API docs.
• Information updates for remaining model cards (also includes the ones in demo, but not in the repository).

Changed ⚠️

• Updated Dockerfile.release and Dockerfile.commit to work with different CUDA versions.
• Changes required for the transformers dependency update to version 4.0.1.

Commits

2c3d00f add missing taskcards (#183)
22cbb2f Update transformers version (#181)
7204469 ci fix (#182)
7c528af Fix typo
780b095 Add check for None before iterating on the clusters. (#179)
ef5d422 Treat auxiliary verbs as verbs (#178)
ea1f71c fix index bug (#175)
8d09185 More model cards (#177)
8a8d8b4 add link to source code in docs (#173)
0789d72 update alternative Dockerfiles for different CUDA versions (#170)
01bc777 actions quick fix

v1.2.2

What's new

Changed ⚠️

• Changed AllenNLP dependency for releases to allow for a range of versions, instead
  of being pinned to an exact version.
• There will now be multiple Docker images pushed to Docker Hub for releases, each
  corresponding to a different supported CUDA version (currently just 10.2 and 11.0).

Fixed ✅

• Fixed pair-classification-esim pretrained model.
• Fixed ValueError error message in Seq2SeqDatasetReader.
• Better check for start and end symbols in Seq2SeqDatasetReader that doesn't fail for BPE-based tokenizers.

Added 🎉

• Added short_description field to ModelCard.
• Information updates for all model cards.

Commits

98ae92c build quickfix
c5e54c9 Prepare for release v1.2.2
04d1e58 Merge branch 'master' of github.com:allenai/allennlp-models
4905da2 install torch in test Docker image separately
716d48d Merge GH Actions workflows into one, publish release images for multiple supported CUDA versions (#168)
1a17c81 More explicit start/end symbol parsing (#169)
236034f Fix bad start or end symbol error (#167)
8a305a0 Update model cards (#166)
43afd10 allow for range of AllenNLP versions (#165)
f2f4284 upload new esim model with fixed config (#164)