GitLab

* move generation_*.py src files into generation/*.py

* populate generation.__init__ with lazy loading

* move imports and references from generation.xxx.object to generation.object

* Use beam search functionality; Add extra outputs and test

* Add full tests for contrastive search

* Add error message on unconventional cache format

* contrastive search test updates

* make fixup

Adding the state-of-the-art contrastive search decoding methods for the codebase of generation_utils.py (#19477)

* add: the contrastive search for generaton_utils

* add: testing scripts for contrastive search under examples/text-generation

* update the quality of codes

* revise the docstring; make the generation_contrastive_search.py scripts;

* revise the examples/pytorch/text-generation/run_generation_contrastive_search.py to the auto-APIs format

* revise the necessary documents

* fix: revise the docstring of generation_contrastive_search.py

* Fix the code indentation

* fix: revise the nits and examples in contrastive_search docstring.

* fix the copyright

* delete generation_contrastive_search.py

* revise the logic in contrastive_search

* update the intergration test and the docstring

* run the tests over

* add the slow decorate to the contrastive_search intergrate test

* add more test

* do the style, quality, consistency checks

* validate generate model_kwargs

* generate tests -- not all models have an attn mask

* fix
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

- Fix `top_k_top_p_filtering` not passing `filter_value` to
   `TopPLogitsWarper` causing any top-p filtered logits to be -inf
   instead of specified value

 - Add corresponding test

* [Draft] fix transition probs

* up

* up

* up

* make it work

* fix

* finish

* update

* Black preview

* Fixup too!

* Fix check copies

* Use the same version as the CI

* Bump black

* min length must be smaller than max length

* Update min_length in tests

* up

* fix min lengths

[Bug Fix] Beam search example in docs fails & a fix (integrating `max_length` in `BeamScorer.finalize()`) (#15555)

* added the test and fix

* had left out a comment

* added classes to get started with constrained beam search

* in progress, think i can directly force tokens now but not yet with the round robin

* think now i have total control, now need to code the bank selection

* technically works as desired, need to optimize and fix design choices leading to undersirable outputs

* complete PR #1 without disjunctive decoding

* removed incorrect tests

* Delete k.txt

* Delete test.py

* Delete test.sh

* revert changes to test scripts

* genutils

* full implementation with testing, no disjunctive yet

* shifted docs

* passing all tests realistically ran locally

* removing accidentally included print statements

* fixed source of error in initial PR test

* fixing the get_device() vs device trap

* fixed documentation docstrings about constrained_beam_search

* fixed tests having failing for Speech2TextModel's floating point inputs

* fix cuda long tensor

* added examples and testing for them and founx & fixed a bug in beam_search and constrained_beam_search

* deleted accidentally added test halting code with assert False

* code reformat

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py

* fixing based on comments on PR

* took out the testing code that should but work fails without the beam search moditification ; style changes

* fixing comments issues

* docstrings for ConstraintListState

* typo in PhrsalConstraint docstring

* docstrings improvements

* finished adding what is sort of an opinionated implementation of disjunctive generation, but it revealed errors in inner beam search logic during testing.

* fixed bug found in constrained beam search that used beam_idx that were not global across all the batches

* disjunctive constraint working 100% correctly

* passing all tests

* Accidentally included mlruns

* Update src/transformers/generation_beam_constraints.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update src/transformers/generation_beam_constraints.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* complete overhaul of type complexities and other nits

* strict type checks in generate()

* fixing second round of feedback by narsil

* fixed failing generation test because of type check overhaul

* generation test fail fix

* fixing test fails
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Per-folder tests reorganization
Co-authored-by: sgugger <sylvain.gugger@gmail.com>
Co-authored-by: Stas Bekman <stas@stason.org>

* added classes to get started with constrained beam search

* in progress, think i can directly force tokens now but not yet with the round robin

* think now i have total control, now need to code the bank selection

* technically works as desired, need to optimize and fix design choices leading to undersirable outputs

* complete PR #1 without disjunctive decoding

* removed incorrect tests

* Delete k.txt

* Delete test.py

* Delete test.sh

* revert changes to test scripts

* genutils

* full implementation with testing, no disjunctive yet

* shifted docs

* passing all tests realistically ran locally

* removing accidentally included print statements

* fixed source of error in initial PR test

* fixing the get_device() vs device trap

* fixed documentation docstrings about constrained_beam_search

* fixed tests having failing for Speech2TextModel's floating point inputs

* fix cuda long tensor

* added examples and testing for them and founx & fixed a bug in beam_search and constrained_beam_search

* deleted accidentally added test halting code with assert False

* code reformat

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* Update tests/test_generation_utils.py

* fixing based on comments on PR

* took out the testing code that should but work fails without the beam search moditification ; style changes

* fixing comments issues

* docstrings for ConstraintListState

* typo in PhrsalConstraint docstring

* docstrings improvements
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* better

* save intermediate

* finish code

* up

* docs

* Apply suggestions from code review

* up

* add compute transition  beam scores function to model and make sure scores are correct with eos

* apply nicos comments

* Apply suggestions from code review

* another fix

* [Generate] correct encoder_outputs are passed without attention_mask

* Apply suggestions from code review

* up

* up

* save

* correct

* up

* correct more

* up

* up

* up

* up

* up

* correct

* fix tf

* fix

* remove tokenizer

* add custom `stopping_criteria` and `logits_processor` to `generate`

* add tests for custom `stopping_criteria` and `logits_processor`

* fix typo in RAG

* address reviewer comments

* improve custom logits processor/stopping criteria error message

* fix types in merge function signature

* change default for custom list from `None` to empty list

* fix rag generate

* add string split suggestion
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>

* [ImageGPT] Deprecate pixel_values input name to input_ids

* up

* Apply suggestions from code review
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* correct

* finish
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>

* [Generate] Correct input_ids detection

* correct

* finish refactor

* refactor

* add tests

* add more tests

* up

* finish tests

* finish

* up

* Apply suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* improve docstring

* fix docs
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* up

* finalize

* finalize

* finish

* Update src/transformers/generation_utils.py

* apply feedback

* up

* Update src/transformers/generation_stopping_criteria.py

* finish

* fix_torch_device_generate_test

* remove @

* fix tests

* 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

* 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_torch_device_generate_test

* remove @

* 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

* init bigbird pegasus

* add debugging nb ; update config

* init conversion

* update conversion script

* complete conversion script

* init forward()

* complete forward()

* add tokenizer

* add some slow tests

* commit current

* fix copies

* add docs

* add conversion script for bigbird-roberta-summarization

* remove TODO

* small fixups

* correct tokenizer

* add bigbird core for now

* fix config

* fix more

* revert pegasus-tokenizer back

* make style

* everything working for pubmed; yayygit status

* complete tests finally

* remove bigbird pegasus tok

* correct tokenizer

* correct tests

* add tokenizer files

* finish make style

* fix test

* update

* make style

* fix tok utils base file

* make fix-copies

* clean a bit

* small update

* fix some suggestions

* add to readme

* fix a bit, clean tests

* fix more tests

* Update src/transformers/__init__.py

* Update src/transformers/__init__.py

* make fix-copies

* complete attn switching, auto-padding left

* make style

* fix auto-padding test

* make style

* fix batched attention tests

* put tolerance at 1e-1 for stand-alone decoder test

* fix docs

* fix tests

* correct slow tokenizer conversion

* Apply suggestions from code review
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* complete remaining suggestions

* fix test
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Suraj Patil <surajp815@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.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>

* 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.

* push

* finish

* finish

* make fix copies

* change name

* [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.