* Laying down building stone for more flexible ONNX export capabilities
* Ability to provide a map of config key to override before exporting.
* Makes it possible to export BART with/without past keys.
* Supports simple mathematical syntax for OnnxVariable.repeated
* Effectively apply value override from onnx config for model
* Supports export with additional features such as with-past for seq2seq
* Store the output path directly in the args for uniform usage across.
* Make BART_ONNX_CONFIG_* constants and fix imports.
* Support BERT model.
* Use tokenizer for more flexibility in defining the inputs of a model.
* Add TODO as remainder to provide the batch/sequence_length as CLI args
* Enable optimizations to be done on the model.
* Enable GPT2 + past
* Improve model validation with outputs containing nested structures
* Enable Roberta
* Enable Albert
* Albert requires opset >= 12
* BERT-like models requires opset >= 12
* Remove double printing.
* Enable XLM-Roberta
* Enable DistilBERT
* Disable optimization by default
* Fix missing setattr when applying optimizer_features
* Add value field to OnnxVariable to define constant input (not from tokenizers)
* Add T5 support.
* Simplify model type retrieval
* Example exporting token_classification pipeline for DistilBERT.
* Refactoring to package `transformers.onnx`
* Solve circular dependency & __main__
* Remove unnecessary imports in `__init__`
* Licences
* Use @Narsil's suggestion to forward the model's configuration to the ONNXConfig to avoid interpolation.
* Onnx export v2 fixes (#12388)
* Tiny fixes
Remove `convert_pytorch` from onnxruntime-less runtimes
Correct reference to model
* Style
* Fix Copied from
* LongFormer ONNX config.
* Removed optimizations
* Remvoe bad merge relicas.
* Remove unused constants.
* Remove some deleted constants from imports.
* Fix unittest to remove usage of PyTorch model for onnx.utils.
* Fix distilbert export
* Enable ONNX export test for supported model.
* Style.
* Fix lint.
* Enable all supported default models.
* GPT2 only has one output
* Fix bad property name when overriding config.
* Added unittests and docstrings.
* Disable with_past tests for now.
* Enable outputs validation for default export.
* Remove graph opt lvls.
* Last commit with on-going past commented.
* Style.
* Disabled `with_past` for now
* Remove unused imports.
* Remove framework argument
* Remove TFPreTrainedModel reference
* Add documentation
* Add onnxruntime tests to CircleCI
* Add test
* Rename `convert_pytorch` to `export`
* Use OrderedDict for dummy inputs
* WIP Wav2Vec2
* Revert "WIP Wav2Vec2"
This reverts commit f665efb04c92525c3530e589029f0ae7afdf603e.
* Style
* Use OrderedDict for I/O
* Style.
* Specify OrderedDict documentation.
* Style :)
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Adding support for `pipeline("automatic-speech-recognition")`.
- Ugly `"config"` choice for AutoModel. It would be great to have the
possibility to have something like `AutoModelFor` that would implement
the same logic (Load the config, check Architectures and load the first
one)
* Remove `model_id` was not needed in the end.
* Rebased !
* Remove old code.
* Rename `nlp`.
* Copy BART to MBart and rename some stuff
* Add copy statements pointing to FlaxBart
* Update/add some common files
* Update shift_tokens_rigth + fix imports
* Fix shift_tokens_right method according to MBart implementation
* Update shift_tokens_right in tests accordingly
* Fix the import issue and update docs file
* make style quality
* Do some minor changes according to patil-suraj suggestions
* Change the order of normalization layer and attention
* Add some copu statementes
* Update generate method and add integration test for mBart
* Make a few updates after a review
Besides, add `lang_code_to_id` to MBartTokenizeFast
* fix-copies; make style quality
* Apply suggestions from code review
* Apply suggestions from code review
* Apply suggestions from code review
* fix output type, style
* add copied from
* resolve conflicts
Co-authored-by: Suraj Patil <surajp815@gmail.com>
* First pass
* More progress
* Add support for local attention
* More improvements
* More improvements
* Conversion script working
* Add CanineTokenizer
* Make style & quality
* First draft of integration test
* Remove decoder test
* Improve tests
* Add documentation
* Mostly docs improvements
* Add CanineTokenizer tests
* Fix most tests on GPU, improve upsampling projection
* Address most comments by @dhgarrette
* Remove decoder logic
* Improve Canine tests, improve docs of CanineConfig
* All tokenizer tests passing
* Make fix-copies and fix tokenizer tests
* Fix test_model_outputs_equivalence test
* Apply suggestions from @sgugger's review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Address some more comments
* Add support for hidden_states and attentions of shallow encoders
* Define custom CanineModelOutputWithPooling, tests pass
* First pass
* More progress
* Add support for local attention
* More improvements
* More improvements
* Conversion script working
* Add CanineTokenizer
* Make style & quality
* First draft of integration test
* Remove decoder test
* Improve tests
* Add documentation
* Mostly docs improvements
* Add CanineTokenizer tests
* Fix most tests on GPU, improve upsampling projection
* Address most comments by @dhgarrette
* Remove decoder logic
* Improve Canine tests, improve docs of CanineConfig
* All tokenizer tests passing
* Make fix-copies and fix tokenizer tests
* Fix test_model_outputs_equivalence test
* Apply suggestions from @sgugger's review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Address some more comments
* Make conversion script work for Canine-c too
* Fix tokenizer tests
* Remove file
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* [WIP] Easily train a new fast tokenizer from a given one
* Fix test
* Roll out to other tokenizers and add tests
* Fix bug with unk id and add emoji to test
* Really use something different in test
* Implement special tokens map
* Map special tokens in the Transformers tokenizers
* Fix test
* Make test more robust
* Fix test for BPE
* More robust map and test
Co-authored-by SaulLu
* Test file
* Stronger tests
Co-authored-by: SaulLu <lucilesaul.com@gmail.com>
* Map unk token for Wordpiece and address review comment
* Fix lowercase test and address review comment
* Fix all tests
* Simplify test
* Fix tests for realsies
* Easily train a new fast tokenizer from a given one - tackle the special tokens format (str or AddedToken) (#12420)
* Propose change in tests regarding lower case
* add new test for special tokens types
* put back the test part about decoding
* add feature: the AddedToken is re-build with the different mapped content
* Address review comment: simplify AddedToken building
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
* Update src/transformers/tokenization_utils_fast.py
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: SaulLu <lucilesaul.com@gmail.com>
Co-authored-by: SaulLu <55560583+SaulLu@users.noreply.github.com>
* Clean push to hub API
* Create working dir if it does not exist
* Different tweak
* New API + all models + test Flax
* Adds the Trainer clean up
* Update src/transformers/file_utils.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Address review comments
* (nit) output types
* No need to set clone_from when folder exists
* Update src/transformers/trainer.py
Co-authored-by: Julien Chaumond <julien@huggingface.co>
* Add generated_from_trainer tag
* Update to new version
* Fixes
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Julien Chaumond <julien@huggingface.co>
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
* copy pytorch-t5
* init
* boom boom
* forward pass same
* make generation work
* add more tests
* make test work
* finish normal tests
* make fix-copies
* finish quality
* correct slow example
* correct slow test
* version table
* upload models
* Update tests/test_modeling_flax_t5.py
* correct incorrectly deleted line
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Patrick von Platen <patrick@huggingface.co>
* Add output args to greedy search
* Fix critical typo + make style quality
* Handle generate_beam_search
* Add dict_specific tests and fix the placement of encoder outputs
* Add specific outputs
* Update doc
* Fix typo
* Adjust handling encoder_outputs + Fix generating for T5
* Fix generate for RAG
* Fix handling ouptut_attentions when target_mapping is not None
Take care of situations when target_mapping is provided
as there are 2-tuple of attentions
Change from:
if inputs["output_attentions"]:
attentions = tuple(tf.transpose(t, perm(2, 3, 0, 1)) for t in attentions)
to:
if inputs["output_attentions"]:
if inputs["target_mapping"] is not None:
# when target_mapping is provided, there are 2-tuple of attentions
attentions = tuple(
tuple(tf.transpose(attn_stream, perm=(2, 3, 0, 1)) for attn_stream in t) for t in attentions
)
else:
attentions = tuple(tf.transpose(t, perm=(2, 3, 0, 1)) for t in attentions)
* Rename kwargs to model_kwargs
* make style quality
* Move imports in test_modeling_tf_common.py
Move ModelOutput-related imports in test_modeling_tf_common.py
into the `is_tf_available():` statement.
* Rewrite nested if-statements
* Fix added tests
* Optimizing away the `fill-mask` pipeline.
- Don't send anything to the tokenizer unless needed. Vocab check is
much faster
- Keep BC by sending data to the tokenizer when needed. User handling warning messages will see performance benefits again
- Make `targets` and `top_k` work together better `top_k` cannot be
higher than `len(targets)` but can be smaller still.
- Actually simplify the `target_ids` in case of duplicate (it can happen
because we're parsing raw strings)
- Removed useless code to fail on empty strings. It works only if empty
string is in first position, moved to ignoring them instead.
- Changed the related tests as only the tests would fail correctly
(having incorrect value in first position)
* Make tests compatible for 2 different vocabs... (at the price of a
warning).
Co-authored-by: @EtaoinWu
* ValueError working globally
* Update src/transformers/pipelines/fill_mask.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* `tokenizer.vocab` -> `tokenizer.get_vocab()` for more compatiblity +
fallback.
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* boom boom
* remove flax clip example
* allow loading head model with base model weights
* add test
* fix imports
* disable save, load test for clip
* add test_save_load_to_base
* AutoTokenizer: infer the class from the tokenizer config if possible
* Add tests
* Update src/transformers/models/auto/tokenization_auto.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* [WIP] Add TFWav2Vec2Model
Work in progress for adding a tensorflow version of Wav2Vec2
* feedback changes
* small fix
* Test Feedback Round 1
* Add SpecAugment and CTC Loss
* correct spec augment mask creation
* docstring and correct copyright
* correct bugs
* remove bogus file
* finish tests correction
* del unnecessary layers
* Update src/transformers/models/wav2vec2/modeling_tf_wav2vec2.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* make style
* correct final bug
* Feedback Changes
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* feature for tokenizer without slow/legacy version
* format
* modify common test
* add tests
* add PreTrainedTokenizerFast to AutoTokenizer
* format
* change tokenizer common test in order to be able to run test without a slow version
* update tokenizer fast test in order to use `rust_tokenizer_class` attribute instead of `tokenizer_class`
* add autokenizer test
* replace `if self.tokenizer_class is not None` with ` if self.tokenizer_class is None`
* remove obsolete change in comment
* Update src/transformers/tokenization_utils_base.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Update src/transformers/tokenization_utils_fast.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* change `get_main_tokenizer` into `get_tokenizers`
* clarify `get_tokenizers` method
* homogenize with `test_slow_tokenizer` and `test_rust_tokenizer`
* add `test_rust_tokenizer = False` to tokenizer which don't define a fast version
* `test_rust_tokenizer = False` for BertJapaneseTokenizer
* `test_rust_tokenizer = False` for BertJapaneseCharacterTokenizationTest
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Start working on FlaxBart
* Create modeling_flax_bart.py
* Write FlaxBartAttention
* Add FlaxBartEncoderLayer
* Add FlaxBartDecoderLayer and some typing
* Add helepr function for FlaxBart
* shift_tokens_right
* _make_causal_mask
* _expand_mask
* Add PositionalEmbedding and fix init_std naming
* Add FlaxBartPretrainedModel
* Add FlaxBartEncoder
* Add FlaxBartEncoder
* Add FlaxBartEncoder among modules to be imported
* YET WE CANNOT INITIALIZE THAT!! :(
* Make BartEncoder working
Change BartEncoder to instance of nn.Module so far
* Add FlaxBartDecoder
* Add FlaxBartModel
* TODO to make model run -> Prepapre model inputs
* Resolve padding
* Add FlaxBartModel
* Add FlaxBartModel into importable modules
* Remove FlaxBartEncoder and FlaxBartDecoder from importable modules
* make style; not properly working
* make style; make quality not pass due to some import I left
* Remove TODO for padding_idx in nn.Embed so far
* Add FlaxBartForConditionalGeneration
* Incorporate Flax model output classes, i.e. return_dict
* Add another models and incorporate use_cache arg
* Add FlaxBartForSequenceClassification and FlaxBartForQuestionAnswering
* Incorporate use_cache arg from PyTorch implementation
* Add all necessary Flax output utils
* Add FlaxBartForCausalLM; not working yet'
* Add minor improvements; still lacks some functionality
* Update docs, src and tests
* Add support of FlaxBart to docs/source
* Fix some bugs in FlaxBart souce code
* Add some neccessary tests for FlaxBart models - jit_compilation not passing
* Fix tests and add test_head_masking
* Fix tests for @jax.jit computation
* Add test_head_masking
* Migrate FlaxBart tests from jax.numpy to numpy
* Remove FlaxBartForCausalLM
* Clean repo
* fix bart model weight structure
* Fix FlaxBartForSequenceClassification
Slicing is not possible to use below jit, therefore, selecting sentence
representation from hidden_states must be changed.
* Allow FlaxBartForSequenceClassification for testing pt_flax equivalence
* Allow testing for FlaxBartForQA for pt_flax equivalence
* Add a comment to FlaxBartForSequenceClassification + change noise from 1e-3 to 1e-6
* remove past_key_values
* remove inputs_mebeds and make input_ids required
* add position ids
* re-write attention layer
* fix dataclass
* fix pos embeds and attention output
* fix pos embeds
* expose encode method
* expose decode method
* move docstring to top
* add cache for causal attn layer
* remove head masking for now
* s2s greedy search first pass
* boom boom
* fix typos
* fix greedy generate for bart
* use encoder, decoder layers instead of num_hidden_layers
* handle encoder_outputs
* cleanup
* simplify decoding
* more clean-up
* typos
* Change header + add {decoder_,}position_ids into 2 models
* add BartConfig
* fix existing tests
* add encode, decode methods
* Fix shift_tokens_right for JIT compilation + clarify one condition
* fix decode
* encoder => encode
* simplify generate
* add tests for encode and decode
* style
* add tests for cache
* fix equivalence tests
* sample generate now works with seq2seq
* generation tests
* initialize dense layers
* docstring and cleanup
* quality
* remove get/set input_embeddings
* address Patricks suggestions
* decode for every model, remove encoder_outputs from call
* update tests accordingly
* decode returns only decoder outputs and logits
* fix arguments
* doc encode, decode methods
* correct base_model_prefix
* fix test for seq classif model
* fix docs
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Suraj Patil <surajp815@gmail.com>
* Fix megatron_gpt2 attention block's causal mask.
* compatibility with checkpoints created with recent versions of Megatron-LM
* added integration test for the released Megatron-GPT2 model
* code style changes
* added option to megatron conversion script to read from config file
Co-authored-by: Guido Novati <gnovati@nvidia.com>
* adding vit for flax
* added test for Flax-vit and some bug-fixes
* overrided methods where variable changes were necessary for flax_vit test
* added FlaxViTForImageClassification for test
* Update src/transformers/models/vit/modeling_flax_vit.py
Co-authored-by: Suraj Patil <surajp815@gmail.com>
* made changes suggested in PR
* Adding jax-vit models for autoimport
* swapping num_channels and height,width dimension
* fixing the docstring for torch-like inputs for VIT
* add model to main init
* add docs
* doc, fix-copies
* docstrings
* small test fixes
* fix docs
* fix docstr
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* style
Co-authored-by: jayendra <jayendra@infocusp.in>
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Fix a condition in test_generate_with_head_masking
* Fix usage of head_mask in bigbirg_pegasus
* Fix head masking for speech2text
* Resolve copy mismatch + drop unwanted print statement
* Fix the condition
* Squash all commits of modeling_detr_v7 branch into one
* Improve docs
* Fix tests
* Style
* Improve docs some more and fix most tests
* Fix slow tests of ViT, DeiT and DETR
* Improve replacement of batch norm
* Restructure timm backbone forward
* Make DetrForSegmentation support any timm backbone
* Fix name of output
* Address most comments by @LysandreJik
* Give better names for variables
* Conditional imports + timm in setup.py
* Address additional comments by @sgugger
* Make style, add require_timm and require_vision to testsé
* Remove train_backbone attribute of DetrConfig, add methods to freeze/unfreeze backbone
* Add png files to fixtures
* Fix type hint
* Add timm to workflows
* Add `BatchNorm2d` to the weight initialization
* Fix retain_grad test
* Replace model checkpoints by Facebook namespace
* Fix name of checkpoint in test
* Add user-friendly message when scipy is not available
* Address most comments by @patrickvonplaten
* Remove return_intermediate_layers attribute of DetrConfig and simplify Joiner
* Better initialization
* Scipy is necessary to get sklearn metrics
* Rename TimmBackbone to DetrTimmConvEncoder and rename DetrJoiner to DetrConvModel
* Make style
* Improve docs and add 2 community notebooks
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
* add test_vocab_size for sentencepiece tok.
* add test_get_vocab for sentencepiece tok.
* add test_convert_token_and_id for sentencepiece tok.
* add test_tokenize_and_convert_tokens_to_string for all tok.
* improve test_tokenize_and_convert_tokens_to_string for sp. tok.
* add common tokenizer integration tests
- for albert
- for barthez
* add tokenizer integration tests to bert gen.
* add most tokenizer integration tests
* fix camembert tokenizer integration test
* add tokenizer integration test to marian
* add tokenizer integration test to reformer
* add typing and doc to tokenizer_integration_test_util
* fix tokenizer integration test of reformer
* improve test_sentencepiece_tokenize_and_convert_tokens_to_string
* empty commit to trigger CI
* fix tokenizer integration test of reformer
* remove code not needed anymore
* empty commit to trigger CI
* empty commit to trigger CI
* initial
* code quality test
* code quality
* added test functions in test_modeling_rag.py and test_retrieval_rag.py to test end2end retreiver
* minor change in test_modeling_rag
* fixed tests
* Update examples/research_projects/rag-end2end-retriever/README.md
typo corrected as suggested by lhoestq
Co-authored-by: Quentin Lhoest <42851186+lhoestq@users.noreply.github.com>
* Update examples/research_projects/rag-end2end-retriever/finetune_rag.py
type change suggested by lhoestq
Co-authored-by: Quentin Lhoest <42851186+lhoestq@users.noreply.github.com>
* Update src/transformers/models/rag/retrieval_rag.py
Adding this change as mentioned by lhoestq.
Co-authored-by: Quentin Lhoest <42851186+lhoestq@users.noreply.github.com>
* completed the minor changes suggested by the reviewers
Co-authored-by: Quentin Lhoest <42851186+lhoestq@users.noreply.github.com>
* Added logic to return attention from flax-bert model and added test cases to check that
* Added new line at the end of file to test_modeling_flax_common.py
* fixing code style
* Fixing Roberta and Elextra models too from cpoying bert
* Added temporary hack to not run test_attention_outputs for FlaxGPT2
* Returning attention weights from GPT2 and changed the tests accordingly.
* last fixes
* bump flax dependency
Co-authored-by: jayendra <jayendra@infocusp.in>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Adding new argument `max_new_tokens` for generate.
This is a proposal to add a new argument `max_new_tokens` to `generate`.
This include a `MaxNewTokensCriteria` that enables callers that don't
know about the token length ahead (like pipelines callers) to manage
more easily the length of their generated output.
* Adding a test for the user warning when both`max_length` and
`max_new_tokens` are used together.
* Removed redundant `no_grad`.
* Fix Bart
* Fix Blenderbot{,_small}
* Fix LED
* Fix Marian
* Fix MBart
* Fix Pegasus
* Fix T5
* Add test for generation with head_mask
* Add a common TF test
* Override a test for the LED model as head masking is not yet properly implemented
* Remove all head_masks from input preparation for LED
* Drop masking for T5 as it needs a bit of refactor
* [Trainer] Report both steps and num samples per second
* Fix batch number
* Update src/transformers/trainer_utils.py
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
* Address review comments
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
* add separator for windows
* fixes test_is_copy_consistent on Windows
* fixing writing encoding issue on extended test (for Windows)
* resolving comments
Cleaner and more scalable implementation of symbolic tracing with torch.fx, and provides support for new architectures:
- ALBERT
- DistilBERT
- MobileBERT
- MegatronBERT
- GPT2
- GPT Neo
Co-authored-by: Michael Benayoun <michael@huggingface.co>
* Add missing head masking for generate() function
* Add head_mask, decoder_head_mask and cross_attn_head_mask
into prepare_inputs_for_generation for generate() function
for multiple encoder-decoder models.
* Add test_genereate_with_head_masking
* [WIP] Update the new test and handle special cases
* make style
* Omit ProphetNet test so far
* make fix-copies
* [TokenClassification] Label realignment for subword aggregation
Tentative to replace https://github.com/huggingface/transformers/pull/11622/files
- Added `AggregationStrategy`
- `ignore_subwords` and `grouped_entities` arguments are now fused
into `aggregation_strategy`. It makes more sense anyway because
`ignore_subwords=True` with `grouped_entities=False` did not have a
meaning anyway.
- Added 2 new ways to aggregate which are MAX, and AVERAGE
- AVERAGE requires a bit more information than the others, for now this
case is slightly specific, we should keep that in mind for future
changes.
- Testing has been modified to reflect new argument, and to check the
correct deprecation and the new aggregation_strategy.
- Put the testing argument and testing results for aggregation_strategy,
close together, so that readers can understand what is supposed to
happen.
- `aggregate` is now only tested on a small model as it does not mean
anything to test it globally for all models.
- Previous tests are unchanged in desired output.
- Added a new test case that showcases better the difference between the
FIRST, MAX and AVERAGE strategies.
* Wrong framework.
* Addressing three issues.
1- Tags might not follow B-, I- convention, so any tag should work now
(assumed as B-TAG)
2- Fixed an issue with average that leads to a substantial code change.
3- The testing suite was not checking for the "index" key for "none"
strategy. This is now fixed.
The issue is that "O" could not be chosen by AVERAGE strategy because
those tokens were filtered out beforehand, so their relative scores were
not counted in the average. Now filtering on
ignore_labels will happen at the very end of the pipeline fixing
that issue.
It's a bit hard to make sure this stays like that because we do
not have a end-to-end test for that behavior
* Formatting.
* Adding formatting to code + cleaner handling of B-, I- tags.
Co-authored-by: Francesco Rubbo <rubbo.francesco@gmail.com>
Co-authored-by: elk-cloner <rezakakhki.rk@gmail.com>
* Typo.
Co-authored-by: Francesco Rubbo <rubbo.francesco@gmail.com>
Co-authored-by: elk-cloner <rezakakhki.rk@gmail.com>
* Add 3D attention mask to T5 model (#9643)
Added code for 3D attention mask in T5 model. Similar to BERT model.
* Add test for 3D attention mask
Added test for 3D attention mask: test_decoder_model_past_with_3d_attn_mask()
3D attention mask of the shape [Batch_size, Seq_length, Seq_length] both for
attention mask and decoder attention mask. Test is passing.
* improve slow class tok usage at xlm rob
* add subword regularization for barthez
* improve barthez tok. test
* fix tokenizer tests
* add subword regularization for camembert
* add subword regularization for deberta v2 tokenizer
* add more doc to deberta v2 tokenizer
* add subword regularization for speech to text tok.
* fix sp_model_kwargs type in speech 2 text tok.
* add subword regularization for M2M100 tok.
* add more concrete type hints
* fix tests for m2m100 and s2t tok.
* add missing Any import
* fix syntax error in m2m100 tok.
* fix unpickle of m2m100 and s2t tok.
* fix test of m2m100 and s2t tok.
* improve unpickle of deberta v2 tok.
* add test for pickle of barthez & camembert
* fix pickle of barthez & camembert
* add test for deberta v2 tok. pickle
* fix m2m100 tok. pickle
* fix s2t tok. pickle
* add subword regularization to albert tok.
* refactor subword reg. test into TokenizerTesterMixin
improve albert tok. test
remove sample argument form albert tok.
check subword reg. using TokenizerTesterMixin
improve tok. tests
improve xlm roberta tok. tests
improve xlm roberta tok. tests
* add subword regularization for big bird t.
* improve xlm roberta tok. test
* add subword regularization for mbart50 tok.
* add subword regularization for pegasus tok.
* add subword regularization for reformer tok.
* add subword regularization for T5 tok.
* fix t5 tok. test formatting
* add subword regularization for xlm_proph. tok.
* add subword regularization for xlnet tok.
* add subword regularization for gert_gen tok.
* add typing to tokenizers
* add typing to xlm rob. tok
* add subword regularization for marian tok.
* add reverse tok. test
* fix marian tok test
* fix marian tok test
* fix casing in tok. tests
* fix style of tok. common test
* fix deberta v2 tok test
* add type annotations to tok. tests
* add type annotations to tok. __init__
* add typing to kokenizer
* add type annotations to tok. __init__
* don't specify the default when it's None
* fix barthez tok. doc
* move sentencepiece tok. tests to TokenizerTesterMixin
* fix unused imports
* fix albert tok. test
* add comment to sentencepiece test options
* fix Any import at big bird tok.
* fix Any import at xlm prophetnet tok.
* empty commit to trigger CI
* Improve docs of DeiT and ViT, add community notebook
* Add gitignore for test_samples
* Add notebook with Trainer
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Autogenerate model cards from the Trainer
* ModelCard deprecated
* Fix test
* Style
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Address review comments
* Quality
* With all metadata
* Metadata
* Post-merge conflict mess
* Data args and all examples
* Default license and languages when possible
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* added fix to decode function. added test to qa pipeline tests
* completed topk docstring
* fixed formatting with black
* applied style_doc to fix line length
* Added Big Bird Fast Tokenizer initial file
* style fixes
* flake fixes
* Added big bird fast tokenizer to init files
* Added big bird fast to Auto tokenization
* fix styles
* minor quality fixes
* Added initial test code
* Fix SpmConverter when precompiled_charsmap doesn't exist
* fixed post processor
* minor style fix
* minor fix input names
* Actually fix identity normalization
* style
* Added token type ids to fast tokenizer
* style
* flake fix
* fix copies
Co-authored-by: Anthony MOI <m.anthony.moi@gmail.com>
* Add the ImageClassificationPipeline
* Code review
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>
* Have `load_image` at the module level
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>
* Set generator in dataloader
* Use generator in all random samplers
* Checkpoint all RNG states
* Final version
* Quality
* Test
* Address review comments
* Quality
* Remove debug util
* Add python and numpy RNGs
* Split states in different files in distributed
* Quality
* local_rank for TPUs
* Only use generator when accepted
* Add test
* Set seed to avoid flakiness
* Make test less flaky
* Quality
* add electra model to flax
* Remove Electra Next Sentence Prediction model added by mistake
* fix parameter sharing and loosen equality threshold
* fix styling issues
* add mistaken removen imports
* fix electra table
* Add FlaxElectra to automodels and fixe docs
* fix issues pointed out the PR
* fix flax electra to comply with latest changes
* remove stale class
* add copied from
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* add flax roberta
* make style
* correct initialiazation
* modify model to save weights
* fix copied from
* fix copied from
* correct some more code
* add more roberta models
* Apply suggestions from code review
* merge from master
* finish
* finish docs
Co-authored-by: Patrick von Platen <patrick@huggingface.co>
* Rebase with master
* Minor bug fix in docs
* Copy files from adding_luke_v2 and improve docs
* change the default value of use_entity_aware_attention to True
* remove word_hidden_states
* fix head models
* fix tests
* fix the conversion script
* add integration tests for the pretrained large model
* improve docstring
* Improve docs, make style
* fix _init_weights for pytorch 1.8
* improve docs
* fix tokenizer to construct entity sequence with [MASK] entity when entities=None
* Make fix-copies
* Make style & quality
* Bug fixes
* Add LukeTokenizer to init
* Address most comments by @patil-suraj and @LysandreJik
* rename _compute_extended_attention_mask to get_extended_attention_mask
* add comments to LukeSelfAttention
* fix the documentation of the tokenizer
* address comments by @patil-suraj, @LysandreJik, and @sgugger
* improve docs
* Make style, quality and fix-copies
* Improve docs
* fix docs
* add "entity_span_classification" task
* update example code for LukeForEntitySpanClassification
* improve docs
* improve docs
* improve the code example in luke.rst
* rename the classification layer in LukeForEntityClassification from typing to classifier
* add bias to the classifier in LukeForEntitySpanClassification
* update docs to use fine-tuned hub models in code examples of the head models
* update the example sentences
* Make style & quality
* Add require_torch to tokenizer tests
* Add require_torch to tokenizer tests
* Address comments by @sgugger and add community notebooks
* Make fix-copies
Co-authored-by: Ikuya Yamada <ikuya@ikuya.net>
* prep for deepspeed==0.3.16
* new version
* too soon
* support and test fp32 mode
* troubleshooting doc start
* workaround no longer needed
* add fp32 doc
* style
* cleanup, add tf32 note
* clarify
* release was made
* Adding `AutomaticSpeechRecognitionPipeline`.
- Because we added everything to enable this pipeline, we probably
should add it to `transformers`.
- This PR tries to limit the scope and focuses only on the pipeline part
(what should go in, and out).
- The tests are very specific for S2T and Wav2vec2 to make sure both
architectures are supported by the pipeline. We don't use the mixin for
tests right now, because that requires more work in the `pipeline`
function (will be done in a follow up PR).
- Unsure about the "helper" function `ffmpeg_read`. It makes a lot of
sense from a user perspective, it does not add any additional
dependencies (as in hard dependency, because users can always use their
own load mechanism). Meanwhile, it feels slightly clunky to have so much
optional preprocessing.
- The pipeline is not done to support streaming audio right now.
Future work:
- Add `automatic-speech-recognition` as a `task`. And add the
FeatureExtractor.from_pretrained within `pipeline` function.
- Add small models within tests
- Add the Mixin to tests.
- Make the logic between ForCTC vs ForConditionalGeneration better.
* Update tests/test_pipelines_automatic_speech_recognition.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Adding docs + main import + type checking + LICENSE.
* Doc style !.
* Fixing TYPE_HINT.
* Specifying waveform shape in the docs.
* Adding asserts + specify in the documentation the shape of the input
np.ndarray.
* Update src/transformers/pipelines/automatic_speech_recognition.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Adding require to tests + move the `feature_extractor` doc.
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* removed max_len
* removed max_length from BeamSearchScorer
* correct max length
* finish
* del vim
* finish & add test
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Add cross_attn_head_mask to BART
* Fix cross_attentions in TFBart-like models
* This commit enables returning of `cross_attentions`
for TFBart-like models
* It also fixes attention head masking in cross-attenion module
* Update TF model templates
* Fix missing , in TF model templates
* Fix typo: congig -> config
* removes the creation of separate config objects and uses the existing ones instead+overwrite resize_token_embeddings from parent class because it is not working for the EncoderDecoderModel
* rollback to current version of the huggingface master branch
* reworked version that ties the encoder and decoder config of the parent encoderdecoder instance
* overwrite of resize_token_embeddings throws an error now
* review comment suggestion
Co-authored-by: Suraj Patil <surajp815@gmail.com>
* implemented warning in case encoderdecoder is created with differing configs of encoderdecoderconfig and decoderconfig or encoderconfig
* added test to avoid diverging configs of wrapper class and wrapped classes
* Update src/transformers/models/encoder_decoder/modeling_encoder_decoder.py
* make style
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Add head_mask & decoder_head_mask + some corrections
* Fix head masking for N-grams
* Enable test_headmasking for encoder and decod
* Fix one typo regarding in modeling_propgetnet.py
* Enable test_headmasking for ProphetNetStandaloneDecoderModelTest
and ProphetNetStandaloneEncoderModelTest in test_modeling_prophetnet.py
* make style
* Fix cross_head_mask
* Fix attention head mask naming
* `cross_head_mask` -> `cross_attn_head_mask`
* `cross_layer_head_mask` -> `cross_attn_layer_head_mask`
* Still need to merge #10605 to master to pass the tests
* Fix cross-attention head mask for Torch BART models
* Fix head masking for cross-attention module for the following
models: BART, Blenderbot, Blenderbot_small, M2M_100, Marian, MBart,
Pegasus
* Enable test_headmasking for M2M_100 model
* Fix cross_head_mask for FSMT, LED and T5
* This commit fixes `head_mask` for cross-attention modules
in the following models: FSMT, LED, T5
* It also contains some smaller changes in doc so that
it is be perfectly clear the shape of `cross_head_mask`
is the same as of `decoder_head_mask`
* Update template
* Fix template for BartForCausalLM
* Fix cross_head_mask for Speech2Text models
* Fix cross_head_mask in templates
* Fix args order in BartForCausalLM template
* Fix doc in BART templates
* Make more explicit naming
* `cross_head_mask` -> `cross_attn_head_mask`
* `cross_layer_head_mask` -> `cross_attn_layer_head_mask`
* Fix doc
* make style quality
* Fix speech2text docstring
* Initial support for upload to hub
* push -> upload
* Fixes + examples
* Fix torchhub test
* Torchhub test I hate you
* push_model_to_hub -> push_to_hub
* Apply mixin to other pretrained models
* Remove ABC inheritance
* Add tests
* Typo
* Run tests
* Install git-lfs
* Change approach
* Add push_to_hub to all
* Staging test suite
* Typo
* Maybe like this?
* More deps
* Cache
* Adapt name
* Quality
* MOAR tests
* Put it in testing_utils
* Docs + torchhub last hope
* Styling
* Wrong method
* Typos
* Update src/transformers/file_utils.py
Co-authored-by: Julien Chaumond <julien@huggingface.co>
* Address review comments
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Julien Chaumond <julien@huggingface.co>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Base move
* Examples reorganization
* Update references
* Put back test data
* Move conftest
* More fixes
* Move test data to test fixtures
* Update path
* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Address review comments and clean
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Removed `max_length` from being mandatory within `generate`.
- Moving on to fully using `StoppingCriteria` for `greedy` and `sample`
modes.
- `max_length` still used for `beam_search` and `group_beam_search`
(Follow up PR)
- Fixes a bug with MaxLengthStoppingCriteria (we should stop as soon a
we hit the max_length, the comparison needs to be or equal, that affects
the tests).
- Added options to use `logits_processor` and `stopping_criteria`
directly within `generate` function (so some users can define their own
`logits_processor` and `stopping_criteria`).
- Modified the backward compat tests to make sure we issue a warning.
* Fix `max_length` argument in `generate`.
* Moving validate to being functional.
- Renamed `smax_length` to `stoppping_max_length`.
* Removing `logits_processor` and `stopping_criteria` from `generate`
arguments.
* Deepcopy.
* Fix global variable name.
* Bulk of the work
* Polish and tests
* Update QA Trainer
* Avoid breaking the predict method
* Deprecation warnings
* Store real eval dataloder
* Get eval dataset reference before wrap
* [WIP] Enabling multilingual models for translation pipelines.
* decoder_input_ids -> forced_bos_token_id
* Improve docstring.
* Rebase
* Fixing 2 bugs
- Type token_ids coming from `_parse_and_tokenize`
- Wrong index from tgt_lang.
* Fixing black version.
* Adding tests for _build_translation_inputs and add them for all
tokenizers.
* Mbart actually puts the lang code at the end.
* Fixing m2m100.
* Adding TF support to `deep_round`.
* Update src/transformers/pipelines/text2text_generation.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Adding one line comment.
* Fixing M2M100 `_build_translation_input_ids`, and fix the call site.
* Fixing tests + deep_round -> nested_simplify
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* First draft of deit
* More improvements
* Remove DeiTTokenizerFast from init
* Conversion script works
* Add DeiT to ViT conversion script
* Add tests, add head model, add support for deit in vit conversion script
* Update model checkpoint names
* Update image_mean and image_std, set resample to bicubic
* Improve docs
* Docs improvements
* Add DeiTForImageClassificationWithTeacher to init
* Address comments by @sgugger
* Improve feature extractors
* Make fix-copies
* Minor fixes
* Address comments by @patil-suraj
* All models uploaded
* Fix tests
* Remove labels argument from DeiTForImageClassificationWithTeacher
* Fix-copies, style and quality
* Fix tests
* Fix typo
* Multiple docs improvements
* More docs fixes
* Add a special tokenizer for CPM model
* make style
* fix
* Add docs
* styles
* cpm doc
* fix ci
* fix the overview
* add test
* make style
* typo
* Custom tokenizer flag
* Add REAMDE.md
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
* AutoFeatureExtractor
* Init and first tests
* Tests
* Damn you gitignore
* Quality
* Defensive test for when not all backends are here
* Use pattern for Speech2Text models
* better names
* add attention mixin
* all slow tests in one class
* make helper methods static so we can test
* add local attention tests
* better names
* doc
* apply review suggestions
* Replace pkg_resources with importlib_metadata
Fixes#10964. The other reason for this change is that pkg_resources has been [deprecated](8fe85c22ce) in favor of importlib_metadata.
* Reduce to a single importlib_metadata import switch
* Trigger CI
Co-authored-by: Stas Bekman <stas@stason.org>
* Documentation about loading a fast tokenizer within Transformers
* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* style
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* closes#10258
* typo
* reworked deberta test
* implemented the comments from BigBird01 regarding sequence pair encoding of deberta
* Update style
* VOCAB_FILES_NAMES is now a oneliner as suggested by @sgugger
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* added #fmt: on as requested by @sgugger
* Style
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Squash all commits into one
* Update ViTFeatureExtractor to use image_utils instead of torchvision
* Remove torchvision and add Pillow
* Small docs improvement
* Address most comments by @sgugger
* Fix tests
* Clean up conversion script
* Pooler first draft
* Fix quality
* Improve conversion script
* Make style and quality
* Make fix-copies
* Minor docs improvements
* Should use fix-copies instead of manual handling
* Revert "Should use fix-copies instead of manual handling"
This reverts commit fd4e591bce.
* Place ViT in alphabetical order
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Replace is_sagemaker_distributed_available
* Merge SageMakerTrainer into Trainer
* Test with shorter condition
* Put back deleted line
* Deprecate SageMakerTrainer and SageMakerTrainingArguments
* Apply suggestions from code review
Co-authored-by: Philipp Schmid <32632186+philschmid@users.noreply.github.com>
Co-authored-by: Philipp Schmid <32632186+philschmid@users.noreply.github.com>
* init
* first working test
* added todo for setup.py
* working test for single node multi node ddp and smd
* added tensorflow single node test
* added directory for pytorch and tensorflow due to different requirements.txt
* added directory for pytorch and tensorflow
* added comment for run_glue until it is available
* added output_dir to it
* smaller dataset to make test running faster
* adjust HP and script
* adjusted parameter for tensorflow
* refactored test scripts
* adjusted make file
* init
* first working test
* added todo for setup.py
* working test for single node multi node ddp and smd
* added tensorflow single node test
* added directory for pytorch and tensorflow due to different requirements.txt
* added directory for pytorch and tensorflow
* added comment for run_glue until it is available
* added output_dir to it
* smaller dataset to make test running faster
* adjust HP and script
* adjusted parameter for tensorflow
* refactored test scripts
* adjusted make file
* updated dlc container
* commented in all tests
* added both ecr images
* added new master branches
* debug
* added new datasets version
* init
* strange rebase bug
* removed changes
* changed min version for tests to work
* updated DLC
* added model parallel test
* removed test files
* removed test files
* tested with ned dlc
* added correct sagemaker sdk version
* adjust DLCs for official one
* reworked tests
* quality
* removed default profile added documentation to it
* added step in release for sagemaker tests
* reverted version for example script removed duplicated script and added install from master to requirements.txt
* removed mistaken .DS_Stores from mac
* fixed tests
* added Sylvains feedback
* make style
* added lysandre's feedback
* Added embeddings layer
* Added layoutlm layers, main model, maskedlm and token classification classes
* Added model classes to tf auto models
* Added model to PT to TF conversion script
* Added model to doc README
* Added tests
* Removed unused imports
* Added layoutlm model, test, and doc for sequence classification, and fix imports in __init__.py
* Made tests pass!
* Fixed typos in imports and docs
* Fixed a typo in embeddings layer
* Removed imports
* Fixed formatting issues, imports, tests
* Added layoutlm layers, main model, maskedlm and token classification classes
* Added model classes to tf auto models
* Added model to PT to TF conversion script
* Removed unused imports
* Added layoutlm model, test, and doc for sequence classification, and fix imports in __init__.py
* Made tests pass!
* Fixed typos in imports and docs
* Removed imports
* Fixed small formatting issues
* Removed duplicates import from main __init__.py
* Chnaged deafult arg to true for adding pooling layer to tf layoutlm
* Fixed formatting issues
* Style
* Added copied from to classes copied from bert
* Fixed doc strings examples to work with layoutlm inputs
* Removed PyTorch reference in doc strings example
* Added integration tests
* Cleaned up initialization file
* Updated model checkpoint identifiers
* Fixed imports
Co-authored-by: Amir Tahmasbi <amir@ehsai.ca>
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
* fix backend tokenizer args override: key mismatch
* no touching the docs
* fix mpnet
* add mpnet to test
* fix test
Co-authored-by: theo <theo@matussie.re>
* Added check to ensure model name passed to from_pretrained and model are the same
* Added test to check from_pretrained throws assert error when passed an incompatiable model name
* Modified assert in from_pretrained with f-strings. Modified test to ensure desired assert message is being generated
* Added check to ensure config and model has model_type
* Fix FlauBERT heads
Co-authored-by: vimarsh chaturvedi <vimarsh chaturvedi>
Co-authored-by: Stas Bekman <stas@stason.org>
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
* Create modeling_flax_eletra with code copied from modeling_flax_bert
* Add ElectraForMaskedLM and ElectraForPretraining
* Add modeling test for Flax electra and fix naming and arg in Flax Electra model
* Add documentation
* Fix code style
* Create modeling_flax_eletra with code copied from modeling_flax_bert
* Add ElectraForMaskedLM and ElectraForPretraining
* Add modeling test for Flax electra and fix naming and arg in Flax Electra model
* Add documentation
* Fix code style
* Fix code quality
* Adjust tol in assert_almost_equal due to very small difference between model output, ranging 0.0010 - 0.0016
* Remove redundant ElectraPooler
* save intermediate
* adapt
* correct bert flax design
* adapt roberta as well
* finish roberta flax
* finish
* apply suggestions
* apply suggestions
Co-authored-by: Chris Nguyen <anhtu2687@gmail.com>
* Added debug prints
* Added config
* Added prints
* Added prints
* Added extra samples to SequentialDistributedSampler
* Added extra samples to SequentialDistributedSampler
Updated SequentialDistributedSampler call
* Added deubg prints
* Removed extra prints
* Making predicitons and labels multiple of batchsize
* updated number of microbatches
* Removed extra prints
* Made start_remainder similar to DistributedSamplerWithLoop
* Minor spacing update
* Added debug prints
Added config
Added prints
Added prints
* Added extra samples to SequentialDistributedSampler
Updated SequentialDistributedSampler call
Added extra samples to SequentialDistributedSampler
Added deubg prints
Removed extra prints
Making predicitons and labels multiple of batchsize
updated number of microbatches
Removed extra prints
Squashing redundant commits
* Made start_remainder similar to DistributedSamplerWithLoop
Minor spacing update
Made start_remainder similar to DistributedSamplerWithLoop
* Test and styling
* Rename test
Co-authored-by: Sylvain Gugger <sylvain.gugger@gmail.com>
* Apply black before checking copies
* Fix for class methods
* Deal with lonely brackets
* Remove debug and add forward changes
* Separate copies and fix test
* Add black as a test dependency
* Tests run on Docker
Co-authored-by: Morgan <funtowiczmo@gmail.com>
* Comments from code review
* Reply to itself
* Dependencies
Co-authored-by: Morgan <funtowiczmo@gmail.com>
* [WIP] Adding new parameter to `generate`: `max_time`.
Generation by tokens number is sometimes a bit clunky because we don't
know how many tokens are good enough or even how many tokens are in
the payload (for pipelines users for instance). This leads to hard
to understand behavior.
This PR proposes a new argument `max_time` which is a float of seconds
for the allowed time for `generate` to run on.
Ideally combinations of `max_tokens=None`, `max_time=2` could be used to
generate as many tokens as possible within time budget.
NB: Another possible approach consists of passing a callback to `generate`
putting the caller in charge of the actual decision of when to stop
generating tokens. It opens the door to 'which args should we pass'
to this callback. It's hard to imagine other use-cases for this
early stopping behavior than time (that are not already covered by
parameters of generate)
* Revamp with StoppingCriteria
* Removing deprecated mentions.
* Forgot arguments to stopping criteria.
* Readding max_length it's not just used as a stopping criteria.
* Default value for `stopping_criteria`.
* Address @patrickvonplaten comments.
- More docstrings
- Actual doc
- Include in global namespace
- Remove TF work.
* Put back `max_length` (deprecation different PR).
* Doc quality.
* Fixing old behavior without `stopping_criteria` but with `max_length`.
Making sure we don't break that in the future.
* Adding more tests for possible inconsistencies between
`max_length` and `stopping_criteria`.
* Fixing the torch imports.
* Allow to pass kwargs to model's from_pretrained when using pipeline.
* Disable the use of past_keys_values for GPT2 when exporting to ONNX.
* style
* Remove comment.
* Appease the documentation gods
* Fix style
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
* Create modeling_tf_dpr.py
* Add TFDPR
* Add back TFPegasus, TFMarian, TFMBart, TFBlenderBot
last commit accidentally deleted these 4 lines, so I recover them back
* Add TFDPR
* Add TFDPR
* clean up some comments, add TF input-style doc string
* Add TFDPR
* Make return_dict=False as default
* Fix return_dict bug (in .from_pretrained)
* Add get_input_embeddings()
* Create test_modeling_tf_dpr.py
The current version is already passed all 27 tests!
Please see the test run at :
https://colab.research.google.com/drive/1czS_m9zy5k-iSJbzA_DP1k1xAAC_sdkf?usp=sharing
* fix quality
* delete init weights
* run fix copies
* fix repo consis
* del config_class, load_tf_weights
They shoud be 'pytorch only'
* add config_class back
after removing it, test failed ... so totally only removing "use_tf_weights = None" on Lysandre suggestion
* newline after .. note::
* import tf, np (Necessary for ModelIntegrationTest)
* slow_test from_pretrained with from_pt=True
At the moment we don't have TF weights (since we don't have official official TF model)
Previously, I did not run slow test, so I missed this bug
* Add simple TFDPRModelIntegrationTest
Note that this is just a test that TF and Pytorch gives approx. the same output.
However, I could not test with the official DPR repo's output yet
* upload correct tf model
* remove position_ids as missing keys
* create modeling_tf_rag
* add tests for tf
* add tf tests
* revert wrong pt commit
* further refactor
* further refactor
* refactor
* Update modeling_tf_rag.py
- input_processing
- fix prepare_input_for_generation (mostly fix generate bug)
- bring back from_pretrained hack in order to test generate
* delete colab pieces of code
* Show case of greedy "generate"
Temporarily change from beam_search test to greedy_search test to show case that TF and PT do get equivalent output.
* cosmetic update
* correct typos
* update
* push some progress
* make easy check
* fix rag save from pretrained
* Update src/transformers/modeling_tf_utils.py
* remove commented out lines
* delete unnecessary lines
* add simple test case for nq_checkpoint
Add nq_checkpoint test to show that current version without hack still fails
* temporarily put ugly hack back again
* Add TFRagSequenceForGeneration!!
* __init__.py , import TFRagSequenceForGeneration
* Add TFRagSequence tests!
* rag init.py - add TFRagSequenceForGeneration
* fix from_pretrained
* fix prepare_inputs_for_generation
* Beam search for RagToken!
* minor clean up
* add tf.cast in TFRagModel
* More tf.cast
* Add all remaining tests (still have issues)
* delete all T5 related
* make style
* fix load weight prefix
* fix bart
* fix return_dict for tf_rag
make all tests pass .. Hooray
* fix some tests
* fix code quality
* fix qualtiy check
* finish tests tf rag
* add tf rag to docs
* remove TFT5 from docstring
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* remove TFT5 from docstring
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Delete outdated comments
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* improve doc strings
* add generative model classes
* fix adjust token logic
* refactor generate for TFRag
* using shape_list, not _get_shape
Co-authored-by: Julien Plu <plu.julien@gmail.com>
* axis=[1]->axis=1
* delete NEED_HELP comment
* improve readability
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* improve readability
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* improve readability
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Indicating model is in a developing state in docstrings
As suggested by Julien
* small last changes
* apply sylvains suggestions
* finish tf rag
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: patrickvonplaten <patrick@huggingface.co>
Co-authored-by: Julien Plu <plu.julien@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* offline mode start
* add specific values
* fix fallback
* add test
* better values check and range
* test that actually works
* document the offline mode
* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* more strict check
* cleaner test
* pt-only test
* style
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Introduce save_strategy training argument
* deprecate EvaluationStrategy
* collapse EvaluationStrategy and LoggingStrategy into a single
IntervalStrategy enum
* modify tests to use modified enum
* convbert conversion test
* fin
* fin
* fin
* clean up tf<->pt conversion
* remove from_pt
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>
* Add check-ops script
* Finish to implement check_tf_ops and start the test
* Make the test mandatory only for BERT
* Update tf_ops folder
* Remove useless classes
* Add the ONNX test for GPT2 and BART
* Add a onnxruntime slow test + better opset flexibility
* Fix test + apply style
* fix tests
* Switch min opset from 12 to 10
* Update src/transformers/file_utils.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Fix GPT2
* Remove extra shape_list usage
* Fix GPT2
* Address Morgan's comments
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Conversion from slow to fast for BPE spm vocabs contained an error.
- There is only 1 test currently (tokenizers + slow) that used the modified path
and it's reformer, which does not contain any ids modification so the
bug was silent for now.
- The real issue is that vocab variable was overloaded by
SentencePieceExtractor, leading to Slow specific vocab oddities to be
completely ignored
- The bug was reported here https://github.com/huggingface/transformers/issues/9518
- Ran the complete tokenization test suite with slow without error
(`RUN_SLOW=1 pytest -sv tests/test_tokenization_*`)
* Remove rebase error.
* Adding the fixture.
* Cleaning up `ConversationalPipeline` to support more than DialoGPT.
Currently ConversationalPipeline was heavily biased towards DialoGPT
,which is the default model for this pipeline.
This PR proposes changes to put back the modifications specific to
DialoGPT into tokenizer-specific behavior wherever possible, by
creating `_build_conversation_input_ids` function that takes
conversation as input, and returns a list of ints corresponding
to the tokens. It feels natural to put here because all models
have probably different strategies to build input_ids from the
full conversation and it's the tokenizer's job to transform strings
into tokens (and vice-versa)
If `_build_conversation_input_ids` is missing, previous behavior is
used so we don't break anything so far (except for blenderbot where it's a fix).
This PR also contains a fix for too long inputs. There used
to be dead code for trying to limit the size of incoming input.
The introduced fixed is that we limit
within `_build_conversation_input_ids` to `tokenizer.model_max_length`.
It corresponds to the intent of the removed dead code and is actually
better because it corresponds to `model_max_length` which is different
from `max_length` (which is a default parameter for `generate`).
- Removed `history` logic from the Conversation as it's not relevant
anymore because tokenization logic has been moved to tokenizer.
And tokenizer cannot save any cache, and conversation cannot know
what is relevant or not.
Also it's not usable from `blenderbot` because the input_ids are
not append only (EOS tokens is always at the end).
- Added `iter_texts` method on `Conversation` because all
the code was literred with some form of this iteration of
past/generated_responses.
* Removing torch mention in types.
* Adding type checking to `_build_conversation_input_ids`.
* Fixing import in strings.
Adding new `encoder_no_repeat_ngram_size` to `generate`.
Blenderbot results seemed off compared to original ParlAI script:
`https://parl.ai/projects/recipes/`. Notably the model seems
to repeat a lot what was said during the conversation.
The actual problem was that `no_repeat_ngram_size` actually applies
to the `encoder_input_ids` but HF's `no_repeat_ngram_size` applies
to the previously generated ids (within the decoder). The history
conversation of blenderbot is within the `encoder` part so that
explains why HF's implementation had the repetitions.
This fix was focused on blenderbot *not* small and added tests
for those because they are quite different in configuration.
This change includes:
- Adding a new EncoderNoRepeatLogitProcessor.
- Adding 1 new arg to `generate` (`encoder_no_repeat_ngram_size`)
- Adding 1 new config parameter `encoder_no_repeat_ngram_size`.
- Adding 2 tests, one for the pipeline (high level, inputs exhibited
repeat behavior, one low level for EncoderNoRepeatLogitProcessor)
- Factored NoRepeatLogitProcessor so that logic could be reused.
Further work:
- Blenderbot conversational pipeline still does not behave correctly
as they way input is prepared within the pipeline is still incorrect
(follow up PR)
- Blenderbot allows the bot to have personas, which is done by
prepending "your personna: XXXX" to the input, this could be explored
too in a follow up PR.
@patrickvonplaten
@LysandreJik
* Update src/transformers/generation_logits_process.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/transformers/generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/transformers/generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Update src/transformers/configuration_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Doc quality.
* Fixing test.
* Last fixes.
* Fixing to account for batch_size.
* Update src/transformers/configuration_utils.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Update src/transformers/generation_utils.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Add {decoder_,}head_mask to LED
* Fix create_custom_forward signatue in encoder
* Add head_mask to longformer
* Add head_mask to longformer to fix dependencies
of LED on Longformer.
* Not working yet
* Add mising one input in longofrmer_modeling.py
* make fix-copies
* change tokenizer requirement
* split line
* Correct typo from list to str
* improve style
* make other function pretty as well
* add comment
* correct typo
* add new test
* pass tests for tok without padding token
* Apply suggestions from code review
* Add {decoder_,}head_mask to fsmt_modeling.py
* Enable test_headmasking and some changes to docs
* Remove test_head_masking flag from fsmt test file
Remove test_head_masking flag from test_modeling_fsmt.py
since test_head_masking is set to be True by default (thus it is redundant to store).
* Merge master and remove test_head_masking = True
* Rebase necessary due to an update of jaxlib
* Remove test_head_masking=True in tests/test_modeling_fsmt.py
as it is redundant.
* Adding a new `return_full_text` parameter to TextGenerationPipeline.
For text-generation, it's sometimes used as prompting text.
In that context, prefixing `generated_text` with the actual input
forces the caller to take an extra step to remove it.
The proposed change adds a new parameter (for backward compatibility).
`return_full_text` that enables the caller to prevent adding the prefix.
* Doc quality.
* Remove redundant test_head_masking = True flags
* Remove all redundant test_head_masking flags in PyTorch test_modeling_* files
* Make test_head_masking = True as a default choice in test_modeling_tf_commong.py
* Remove all redundant test_head_masking flags in TensorFlow
test_modeling_tf_* files
* Put back test_head_masking=False fot TFT5 models
* fix --lr_scheduler_type choices
* rewrite to fix for all enum-based cl args
* cleanup
* adjust test
* style
* Proposal that should work
* Remove needless code
* Fix test
Co-authored-by: Sylvain Gugger <sylvain.gugger@gmail.com>
pipeline.
- If table is empty then the line that contain `answer[0]` will fail.
- This PR add a check to prevent `answer[0]`.
- Also adds an early check for presence of `table` and `query` to
prevent late failure and give better error message.
- Adds a few tests to make sure these errors are correctly raised.
* We most likely don't want special tokens in this output.
* Adding `skip_special_tokens=True` to FillMaskPipeline
- It's backward incompatible.
- It makes for sense for pipelines to remove references to
special_tokens (all of the other pipelines do that).
- Keeping special tokens makes it hard for users to actually remove them
because all models have different tokens (<s>, <cls>, [CLS], ....)
* Fixing `token_str` in the same vein, and actually fix the tests too !
* Add head_mask/decoder_head_mask for TF BART models
* Add head_mask and decoder_head_mask input arguments for TF BART-based
models as a TF counterpart to the PR #9569
* Add test_headmasking functionality to tests/test_modeling_tf_common.py
* TODO: Add a test to verify that we can get a gradient back for
importance score computation
* Remove redundant #TODO note
Remove redundant #TODO note from tests/test_modeling_tf_common.py
* Fix assertions
* Make style
* Fix ...Model input args and adjust one new test
* Add back head_mask and decoder_head_mask to BART-based ...Model
after the last commit
* Remove head_mask ande decoder_head_mask from input_dict
in TF test_train_pipeline_custom_model as these two have different
shape than other input args (Necessary for passing this test)
* Revert adding global_rng in test_modeling_tf_common.py
* Add decoder_head_mask for PyTorch T5 model
* Add decoder_head_mask args into T5Model and T5ForConditionalGeneration
* Slightly change the order of input args to be in accordance
with the convention from BART-based models introduced within the PR #9569.
* Make style for modeling_t5.py
* Add decoder_head_mask for TF T5 models
* Separate head_mask and decoder_head_mask args in TF T5 models
* Slightly change the order of input args to follow convention
of BART-based models updated in PR #9569
* Update test_forward_signature tests/test_modeling_tf_common.py
w.r.t. the changed order of input args
* Add FutureWarnings for T5 and TFT5 models
* Add FutureWarnings for T5 and TFT5 models warning a user that
input argument `head_mask` was split into two arguments -
`head_mask` and `decoder_head_mask`
* Add default behaviour - `decoder_head_mask` is set to copy
`head_mask`
* Fix T5 modeling and FutureWarning
* Make proper usage of head_mask and decoder_head_mask
in cross_attention
* Fix conditions for raising FutureWarning
* Reformat FutureWarning in T5 modeling
* Refactor the warning message
* Update past_key_values in gpt2 (#9391)
* Update generation_utils, and rename some items
* Update modeling_gpt2 to avoid an error in gradient_checkpointing
* Remove 'reorder_cache' from util and add variations to XLNet, TransfoXL, GPT-2
* Change the location of '_reorder_cache' in modeling files
* Add '_reorder_cache' in modeling_ctrl
* Fix a bug of my last commit in CTRL
* Add '_reorder_cache' to GPT2DoubleHeadsModel
* Manage 'use_cache' in config of test_modeling_gpt2
* Clean up the doc string
* Update src/transformers/models/gpt2/modeling_gpt2.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Fix the doc string (GPT-2, CTRL)
* improve gradient_checkpointing_behavior
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Add head_mask/decoder_head_mask for BART
This branch implement head_mask and decoder_head_mask
for BART-based models. Full list below:
- BART
- MBart
- Blenderbot
- BlenderbotSmall
- Marian
- Pegasus
Everything is accompanied with updated testing.
* Fix test_headmasking for BART models
* Fix text_headmasking for BART-like models
which has only 2 layers in each modules.
The condition
```
self.assertNotEqual(attentions[1][..., 0, :, :].flatten().sum().item(), 0.0)
```
is, therefore, invalid for encoder-decoder models considering
the `head_mask`
```
head_mask = torch.ones(
self.model_tester.num_hidden_layers,
self.model_tester.num_attention_heads,
device=torch_device,
)
head_mask[0, 0] = 0
head_mask[-1, :-1] = 0
```
specified in the `test_headmasking` test/function.
* Adjust test_modeling_common.py to reflect T5 input args
* Update tests/test_modeling_common.py
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* make style
* make fix-copies
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Add target contextmanager and rework prepare_seq2seq_batch
* Fix tests, treat BART and Barthez
* Add last tokenizers
* Fix test
* Set src token before calling the superclass
* Remove special behavior for T5
* Remove needless imports
* Remove needless asserts
* Add LayoutLMForSequenceClassification and integration tests
Improve docs
Add LayoutLM notebook to list of community notebooks
* Make style & quality
* Address comments by @sgugger, @patrickvonplaten and @LysandreJik
* Fix rebase with master
* Reformat in one line
* Improve code examples as requested by @patrickvonplaten
Co-authored-by: Lysandre <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Enable TruncationStrategy override for pipelines
* Update isort.
* Fixing test
* Fixing text_generation pipeline.
* Using same DummyTok as other PR for easier merge later.
* Some more import guards.
* Remove bogus file.
* Do not pass `generate_kwargs` to `_parse_and_tokenize`.
@patrickvonplaten
* Removed DummyTok.
* Doc quality.
* Cleaning up conversation tests.
* Adding tests that don't require downloading models + conversation can be
fully created from static state.
* Making tests non flaky (by fixing generation length)
* Bumping isort version.
* Doc cleanup.
* Remove unused test in this PR.
* Torch import guard for TF.
* Missing torch guard.
* Small mistake in doc.
* Actual uses `_history` and `_index` cache.
+ remove dead enumerate
+ improve warning message.
* Update src/transformers/pipelines/conversational.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Update src/transformers/pipelines/conversational.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Update src/transformers/pipelines/conversational.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Adding comments and cleaner code to address history copy.
* Improving pipeline name in tests.
* Change tokenizer to a real one (still created at runtime with no
external dependency)
* Simplify DummyTok, reverse changes on tokenization.
* Removing DummyTok.
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Don't import libs to check they are available
* Don't import integrations at init
* Add importlib_metdata to deps
* Remove old vars references
* Avoid syntax error
* Adapt testing utils
* Try to appease torchhub
* Add dependency
* Remove more private variables
* Fix typo
* Another typo
* Refine the tf availability test
* Define new output dataclasses for greedy generation
* Add output_[...] flags in greedy generation methods
Added output_attentions, output_hidden_states, output_scores flags in
generate and greedy_search methods in GenerationMixin.
* [WIP] Implement logic and tests for output flags in generation
* Update GreedySearchOutput classes & docstring
* Implement greedy search output accumulation logic
Update greedy_search unittests
Fix generate method return value docstring
Properly init flags with the default config
* Update configuration to add output_scores flag
* Fix test_generation_utils
Sort imports and fix isinstance tests for GreedySearchOutputs
* Fix typo in generation_utils
* Add return_dict_in_generate for backwards compatibility
* Add return_dict_in_generate flag in config
* Fix tyPo in configuration
* Fix handling of attentions and hidden_states flags
* Make style & quality
* first attempt attentions
* some corrections
* improve tests
* special models requires special test
* disable xlm test for now
* clean tests
* fix for tf
* isort
* Add output dataclasses for other generation methods
* Add logic to return dict in sample generation
* Complete test for sample generation
- Pass output_attentions and output_hidden_states flags to encoder in
encoder-decoder models
- Fix import satements order in test_generation_utils file
* Add logic to return dict in sample generation
- Refactor tests to avoid using self.assertTrue, which provides
scarce information when the test fails
- Add tests for the three beam_search methods: vanilla, sample and
grouped
* Style doc
* Fix copy-paste error in generation tests
* Rename logits to scores and refactor
* Refactor group_beam_search for consistency
* make style
* add sequences_scores
* fix all tests
* add docs
* fix beam search finalize test
* correct docstring
* clean some files
* Made suggested changes to the documentation
* Style doc ?
* Style doc using the Python util
* Update src/transformers/generation_utils.py
* fix empty lines
* fix all test
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* create model
* add integration
* save current state
* make integration tests pass
* add one more test
* add explanation to tests
* remove from bart
* add padding
* remove unnecessary test
* make all tests pass
* re-add cookie cutter tests
* finish PyTorch
* fix attention test
* Update tests/test_modeling_common.py
* revert change
* remove unused file
* add string to doc
* save intermediate
* make tf integration tests pass
* finish tf
* fix doc
* fix docs again
* add led to doctree
* add to auto tokenizer
* added tips for led
* make style
* apply jplus statements
* correct tf longformer
* apply lysandres suggestions
* apply sylvains suggestions
* Apply suggestions from code review
* Create modeling_tf_dpr.py
* Add TFDPR
* Add back TFPegasus, TFMarian, TFMBart, TFBlenderBot
last commit accidentally deleted these 4 lines, so I recover them back
* Add TFDPR
* Add TFDPR
* clean up some comments, add TF input-style doc string
* Add TFDPR
* Make return_dict=False as default
* Fix return_dict bug (in .from_pretrained)
* Add get_input_embeddings()
* Create test_modeling_tf_dpr.py
The current version is already passed all 27 tests!
Please see the test run at :
https://colab.research.google.com/drive/1czS_m9zy5k-iSJbzA_DP1k1xAAC_sdkf?usp=sharing
* fix quality
* delete init weights
* run fix copies
* fix repo consis
* del config_class, load_tf_weights
They shoud be 'pytorch only'
* add config_class back
after removing it, test failed ... so totally only removing "use_tf_weights = None" on Lysandre suggestion
* newline after .. note::
* import tf, np (Necessary for ModelIntegrationTest)
* slow_test from_pretrained with from_pt=True
At the moment we don't have TF weights (since we don't have official official TF model)
Previously, I did not run slow test, so I missed this bug
* Add simple TFDPRModelIntegrationTest
Note that this is just a test that TF and Pytorch gives approx. the same output.
However, I could not test with the official DPR repo's output yet
* upload correct tf model
* remove position_ids as missing keys
* fix RagSeq generate with context_input_ids
fix RagSeq generate with context_input_ids
* apply style
* delete unused lines
* Add test_rag_sequence_generate_batch_from_context_input_ids
* Readability improved
* stylying
* Stylize
* typos
* add check_model_generate_from_context_input_ids
* make style
* Apply suggestions from code review
* make style2
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: patrickvonplaten <patrick@huggingface.co>
* add past_key_values
* add use_cache option
* make mask before cutting ids
* adjust position_ids according to past_key_values
* flatten past_key_values
* fix positional embeds
* fix _reorder_cache
* set use_cache to false when not decoder, fix attention mask init
* add test for caching
* add past_key_values for Roberta
* fix position embeds
* add caching test for roberta
* add doc
* make style
* doc, fix attention mask, test
* small fixes
* adress patrick's comments
* input_ids shouldn't start with pad token
* use_cache only when decoder
* make consistent with bert
* make copies consistent
* add use_cache to encoder
* add past_key_values to tapas attention
* apply suggestions from code review
* make coppies consistent
* add attn mask in tests
* remove copied from longformer
* apply suggestions from code review
* fix bart test
* nit
* simplify model outputs
* fix doc
* fix output ordering
* Add label smoothing in Trainer
* Add options for scheduler and Adafactor in Trainer
* Put Seq2SeqTrainer in the main lib
* Apply suggestions from code review
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Address review comments and adapt scripts
* Documentation
* Move test not using script to tests folder
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Add new run_swag example
* Add check
* Add sample
* Apply suggestions from code review
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* Very important change to make Lysandre happy
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
* First commit: adding all files from tapas_v3
* Fix multiple bugs including soft dependency and new structure of the library
* Improve testing by adding torch_device to inputs and adding dependency on scatter
* Use Python 3 inheritance rather than Python 2
* First draft model cards of base sized models
* Remove model cards as they are already on the hub
* Fix multiple bugs with integration tests
* All model integration tests pass
* Remove print statement
* Add test for convert_logits_to_predictions method of TapasTokenizer
* Incorporate suggestions by Google authors
* Fix remaining tests
* Change position embeddings sizes to 512 instead of 1024
* Comment out positional embedding sizes
* Update PRETRAINED_VOCAB_FILES_MAP and PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
* Added more model names
* Fix truncation when no max length is specified
* Disable torchscript test
* Make style & make quality
* Quality
* Address CI needs
* Test the Masked LM model
* Fix the masked LM model
* Truncate when overflowing
* More much needed docs improvements
* Fix some URLs
* Some more docs improvements
* Test PyTorch scatter
* Set to slow + minify
* Calm flake8 down
* First commit: adding all files from tapas_v3
* Fix multiple bugs including soft dependency and new structure of the library
* Improve testing by adding torch_device to inputs and adding dependency on scatter
* Use Python 3 inheritance rather than Python 2
* First draft model cards of base sized models
* Remove model cards as they are already on the hub
* Fix multiple bugs with integration tests
* All model integration tests pass
* Remove print statement
* Add test for convert_logits_to_predictions method of TapasTokenizer
* Incorporate suggestions by Google authors
* Fix remaining tests
* Change position embeddings sizes to 512 instead of 1024
* Comment out positional embedding sizes
* Update PRETRAINED_VOCAB_FILES_MAP and PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
* Added more model names
* Fix truncation when no max length is specified
* Disable torchscript test
* Make style & make quality
* Quality
* Address CI needs
* Test the Masked LM model
* Fix the masked LM model
* Truncate when overflowing
* More much needed docs improvements
* Fix some URLs
* Some more docs improvements
* Add add_pooling_layer argument to TapasModel
Fix comments by @sgugger and @patrickvonplaten
* Fix issue in docs + fix style and quality
* Clean up conversion script and add task parameter to TapasConfig
* Revert the task parameter of TapasConfig
Some minor fixes
* Improve conversion script and add test for absolute position embeddings
* Improve conversion script and add test for absolute position embeddings
* Fix bug with reset_position_index_per_cell arg of the conversion cli
* Add notebooks to the examples directory and fix style and quality
* Apply suggestions from code review
* Move from `nielsr/` to `google/` namespace
* Apply Sylvain's comments
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Co-authored-by: Rogge Niels <niels.rogge@howest.be>
Co-authored-by: LysandreJik <lysandre.debut@reseau.eseo.fr>
Co-authored-by: Lysandre Debut <lysandre@huggingface.co>
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
* add model parallelism to T5EncoderModel
add model parallelism to T5EncoderModel
* remove decoder from T5EncoderModel parallelize
* uodate T5EncoderModel docs
* Extend T5ModelTest for T5EncoderModel
* fix T5Stask using range for get_device_map
* fix style
Co-authored-by: Ahmed Elnaggar <elnaggar@rostlab.informatik.tu-muenchen.de>
* Resize the biases in same time than the embeddings
* Trigger CI
* Biases are not reset anymore
* Remove get_output_embeddings + better LM model detection in generation utils
* Apply style
* First test on BERT
* Update docstring + new name
* Apply the new resizing logic to all the models
* fix tests
* Apply style
* Update the template
* Fix naming
* Fix naming
* Apply style
* Apply style
* Remove unused import
* Revert get_output_embeddings
* Trigger CI
* Update num parameters
* Restore get_output_embeddings in TFPretrainedModel and add comments
* Style
* Add decoder resizing
* Style
* Fix tests
* Separate bias and decoder resize
* Fix tests
* Fix tests
* Apply style
* Add bias resizing in MPNet
* Trigger CI
* Apply style
* remove make on the fly linear embedding
* start refactor
* big first refactor
* save intermediate
* save intermediat
* correct mask issue
* save tests
* refactor padding masks
* make all tests pass
* further refactor
* make pegasus test pass
* fix bool if
* fix leftover tests
* continue
* bart renaming
* delete torchscript test hack
* fix imports in tests
* correct shift
* fix docs and repo cons
* re-add fix for FSTM
* typo in test
* fix typo
* fix another typo
* continue
* hot fix 2 for tf
* small fixes
* refactor types linting
* continue
* finish refactor
* fix import in tests
* better bart names
* further refactor and add test
* delete hack
* apply sylvains and lysandres commens
* small perf improv
* further perf improv
* improv perf
* fix typo
* make style
* small perf improv
* Remove "Model" suffix from Flax models to look more 🤗
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Initial working (forward + backward) for Flax MLM training example.
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Simply code
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Addressing comments, using module and moving to LM task.
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Restore parameter name "module" wrongly renamed model.
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Restore correct output ordering...
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Actually commit the example 😅
Signed-off-by: Morgan Funtowicz <morgan@huggingface.co>
* Add FlaxBertModelForMaskedLM after rebasing.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Make it possible to initialize the training from scratch
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Reuse flax linen example of cross entropy loss
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Added specific data collator for flax
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Remove todo for data collator
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Added evaluation step
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Added ability to provide dtype to support bfloat16 on TPU
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Enable flax tensorboard output
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Enable jax.pmap support.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Ensure batches are correctly sized to be dispatched with jax.pmap
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Enable bfloat16 with --fp16 cmdline args
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Correctly export metrics to tensorboard
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Added dropout and ability to use it.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Effectively enable & disable during training and evaluation steps.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Oops.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Enable specifying kernel initializer scale
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Style.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Added warmup step to the learning rate scheduler.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Fix typo.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Print training loss
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Make style
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* fix linter issue (flake8)
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Fix model matching
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Fix dummies
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Fix non default dtype on Flax models
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Use the same create_position_ids_from_input_ids for FlaxRoberta
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Make Roberta attention as Bert
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* fix copy
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Wording.
Co-authored-by: Marc van Zee <marcvanzee@gmail.com>
Co-authored-by: Marc van Zee <marcvanzee@gmail.com>
* diverse beam search
* bug fixes
* bug fixes
* bug fix
* separate out diverse_beam_search function
* separate out diverse_beam_search function
* bug fix
* improve code quality
* bug fix
* bug fix
* separate out diverse beam search scorer
* code format
* code format
* code format
* code format
* add test
* code format
* documentation changes
* code quality
* add slow integration tests
* more general name
* refactor into logits processor
* add test
* avoid too much copy paste
* refactor
* add to docs
* fix-copies
* bug fix
* Revert "bug fix"
This reverts commit c99eb5a8dc.
* improve comment
* implement sylvains feedback
Co-authored-by: Ayush Jain <a.jain@sprinklr.com>
Co-authored-by: ayushtiku5 <40797286+ayushtiku5@users.noreply.github.com>
* Add new SQUAD example
* Same with a task-specific Trainer
* Address review comment.
* Small fixes
* Initial work for XLNet
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Final clean up and working XLNet script
* Test and debug
* Final working version
* Add new SQUAD example
* Same with a task-specific Trainer
* Address review comment.
* Small fixes
* Initial work for XLNet
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Final clean up and working XLNet script
* Test and debug
* Final working version
* Add tick
* Update README
* Address review comments
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Removed unused `encoder_hidden_states` and `encoder_attention_mask` from MobileBert
* Removed decoder tests for MobileBert
* Removed now unnecessary import
* initial commit
* [cli] lfs commands
* Fix FileSlice
* Tweak to FileSlice
* [hf_api] Backport filetype arg from `datasets`
cc @lhoestq
* Silm down the CI while i'm working
* Ok let's try this in CI
* Update config.yml
* Do not try this at home
* one more try
* Update lfs.py
* Revert "Tweak to FileSlice"
This reverts commit d7e32c4b35.
* Update test_hf_api.py
* Update test_hf_api.py
* Update test_hf_api.py
* CI still green?
* make CI green again?
* Update test_hf_api.py
* make CI red again?
* Update test_hf_api.py
* add CI style back
* Fix CI?
* oh my
* doc + switch back to real staging endpoint
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Pierric Cistac <Pierrci@users.noreply.github.com>
* Fix docblock + f-strings
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Pierric Cistac <Pierrci@users.noreply.github.com>
* Add TFGPT2ForSequenceClassification based on DialogRPT
* Add TFGPT2ForSequenceClassification based on DialogRPT
* TFGPT2ForSequenceClassification based on DialogRPT-refactored code, implemented review comments and added input processing
* Add TFGPT2ForSequenceClassification based on DialogRPT
* TFGPT2ForSequenceClassification based on DialogRPT-refactored code, implemented review comments and added input processing
* code refactor for latest other TF PR
* code refactor
* code refactor
* Update modeling_tf_gpt2.py
* Warning about too long input for fast tokenizers too
If truncation is not set in tokenizers, but the tokenization is too long
for the model (`model_max_length`), we used to trigger a warning that
The input would probably fail (which it most likely will).
This PR re-enables the warning for fast tokenizers too and uses common
code for the trigger to make sure it's consistent across.
* Checking for pair of inputs too.
* Making the function private and adding it's doc.
* Remove formatting ?? in odd place.
* Missed uppercase.
* NerPipeline (TokenClassification) now outputs offsets of words
- It happens that the offsets are missing, it forces the user to pattern
match the "word" from his input, which is not always feasible.
For instance if a sentence contains the same word twice, then there
is no way to know which is which.
- This PR proposes to fix that by outputting 2 new keys for this
pipelines outputs, "start" and "end", which correspond to the string
offsets of the word. That means that we should always have the
invariant:
```python
input[entity["start"]: entity["end"]] == entity["entity_group"]
# or entity["entity"] if not grouped
```
* Fixing doc style
* Slightly increase tolerance between pytorch and flax output
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* test_multiple_sentences doesn't require torch
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Simplify parameterization on "jit" to use boolean rather than str
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Use `require_torch` on `test_multiple_sentences` because we pull the weight from the hub.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Rename "jit" parameter to "use_jit" for (hopefully) making it self-documenting.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Remove pytest.mark.parametrize which seems to fail in some circumstances
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Fix unused imports.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Fix style.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Give default parameters values for traced model.
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Review comment: Change sentences to sequences
Signed-off-by: Morgan Funtowicz <funtowiczmo@gmail.com>
* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Fix decoder not returning hidden states from the last layer
* Resolve conflict
* Change the way to gather hidden states
* Add decoder hidden states test
* Make pytest and black happy
* Remove redundant line
* remove new line
Co-authored-by: Stas Bekman <stas00@users.noreply.github.com>
* fix mems in xlnet
* fix use_mems
* fix use_mem_len
* fix use mems
* clean docs
* fix tf typo
* make xlnet tf for generation work
* fix tf test
* refactor use cache
* add use cache for missing models
* correct use_cache in generate
* correct use cache in tf generate
* fix tf
* correct getattr typo
* make sylvain happy
* change in docs as well
* do not apply to cookie cutter statements
* fix tf test
* make pytorch model fully backward compatible
* bart output hidden states upstream
* same w/ decoder
* add tests
* fix prophetnet
* fix gpt2 and ctrl
* fix fstm and skip test for reformer and longformer
* fix all models
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* implement support for run-time dependency version checking
* try not escaping !
* use findall that works on py36
* small tweaks
* autoformatter worship
* simplify
* shorter names
* add support for non-versioned checks
* add deps
* revert
* tokenizers not required, check version only if installed
* make a proper distutils cmd and add make target
* tqdm must be checked before tokenizers
* workaround the DistributionNotFound peculiar setup
* handle the rest of packages in setup.py
* fully sync setup.py's install_requires - to check them all
* nit
* make install_requires more readable
* typo
* Update setup.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* restyle
* add types
* simplify
* simplify2
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Support BERT relative position embeddings
* Fix typo in README.md
* Address review comment
* Fix failing tests
* [tiny] Fix style_doc.py check by adding an empty line to configuration_bert.py
* make fix copies
* fix configs of electra and albert and fix longformer
* remove copy statement from longformer
* fix albert
* fix electra
* Add bert variants forward tests for various position embeddings
* [tiny] Fix style for test_modeling_bert.py
* improve docstring
* [tiny] improve docstring and remove unnecessary dependency
* [tiny] Remove unused import
* re-add to ALBERT
* make embeddings work for ALBERT
* add test for albert
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Add early stopping patience and minimum threshold metric must improve to prevent early stopping to pytorch trainer
* Add early stopping test
* Set patience counter to 0 if best metric not defined yet
* Make early stopping a callback. Add callback event for updating the best metric for early stopping callback to trigger on.
* Run make style
* make funciton name sensible
* Improve new argument docstring wording and hope that flakey CI test passes.
* Use on_evaluation callback instead of custom. Remove some debug printing
* Move early stopping arguments and state into early stopping callback
* Run make style
* Remove old code
* Fix docs formatting. make style went rogue on me.
* Remove copied attributes and fix variable
* Add assertions on training arguments instead of mutating them. Move comment out of public docs.
* Make separate test for early stopping callback. Add test of invalid arguments.
* Run make style... I remembered before CI this time!
* appease flake8
* Add EarlyStoppingCallback to callback docs
* Make docstring EarlyStoppingCallabck match other callbacks.
* Fix typo in docs
* Make ci fail
* Try to make tests actually run?
* CI finally failing?
* Fix CI
* Revert "Fix CI"
This reverts commit ca7923be73.
* Ooops wrong one
* one more try
* Ok ok let's move this elsewhere
* Alternative to globals() (#8667)
* Alternative to globals()
* Error is raised later so return None
* Sentencepiece not installed make some tokenizers None
* Apply Lysandre wisdom
* Slightly clearer comment?
cc @sgugger
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* `disable_ngram_loss` fix for prophetnet
* add changes documentation
* fix _compute_loss to use mean reduction and -100 to masked tokens & remove unnecessary arguments
* mean label smoothing loss
* small refactor
* fix test
Co-authored-by: patrickvonplaten <patrick.v.platen@gmail.com>
* working on LongformerForSequenceClassification
* add TFLongformerForMultipleChoice
* add TFLongformerForTokenClassification
* use add_start_docstrings_to_model_forward
* test TFLongformerForSequenceClassification
* test TFLongformerForMultipleChoice
* test TFLongformerForTokenClassification
* remove test from repo
* add test and doc for TFLongformerForSequenceClassification, TFLongformerForTokenClassification, TFLongformerForMultipleChoice
* add requested classes to modeling_tf_auto.py
update dummy_tf_objects
fix tests
fix bugs in requested classes
* pass all tests except test_inputs_embeds
* sync with master
* pass all tests except test_inputs_embeds
* pass all tests
* pass all tests
* work on test_inputs_embeds
* fix style and quality
* make multi choice work
* fix TFLongformerForTokenClassification signature
* fix TFLongformerForMultipleChoice, TFLongformerForSequenceClassification signature
* fix mult choice
* fix mc hint
* fix input embeds
* fix input embeds
* refactor input embeds
* fix copy issue
* apply sylvains changes and clean more
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
