* First draft
* More improvements
* More improvements
* Make all tests pass
* Remove script
* Update image processor
* Address comments
* Use new gradient checkpointing method
* Convert checkpoints, add integration test
* Do not keep aspect ratio for now
* Set keep_aspect_ratio=False for beit, add integration test
* Remove print statement
* fixes: code fixes on is_batched condition to also check for batched audio data in torch.Tensor format instead of only just checking for batched audio data in np.ndarray format
* Update src/transformers/models/seamless_m4t/feature_extraction_seamless_m4t.py
Co-authored-by: Yoach Lacombe <52246514+ylacombe@users.noreply.github.com>
* refactor: code refactoring to remove torch framework dependency
* docs: updated docstring to add torch tensor compatibility
* test: add test cases to incorporate torch tensor inputs
* test: ran make fix-copies for code conformity
* test: refactor test to separate the test_call into test_call_numpy and test_call_torch
---------
Co-authored-by: Yoach Lacombe <52246514+ylacombe@users.noreply.github.com>
* Fix vision text dual encoder
* Small cleanup for wav2vec2 (not fixed yet)
* Small fix for vision_encoder_decoder
* Fix SAM builds
* Update TFBertTokenizer test with modern exporting + tokenizer
* Fix DeBERTa
* Fix DeBERTav2
* Try RAG fix but it's impossible to test locally
* Actually fix RAG now that I got FAISS working somehow
* Fix Wav2Vec2, add sermon
* Fix Hubert
* some nits
* update test
* add support d\sd[a
* remove some dummy inputs
* all good
* style
* nits
* fixes
* fix more copies
* nits
* styling
* fix
* Update src/transformers/models/mistral/modeling_mistral.py
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
* add a slow test just to be sure
* fixup
---------
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
* Iteratre over out_features instead of stage_names
* Update for all backbones
* Add tests
* Fix
* Align timm backbone behaviour with other backbones
* Fix tests
* Stricter checks on set out_features and out_indices
* Revert back stage selection logic
* Remove out-of-order logic
* Document restriction in docstrings
* move code to Trainer.evaluate to enable use of that function with multiple datasets
* test
* update doc string
* and a tip
* forgot the type
---------
Co-authored-by: Prof. Peter Schneider-Kamp <jps@ordbogen.com>
* edits to _prepare_4d_causal_attention_mask()
* initial tests for 4d mask
* attention_mask_for_sdpa support
* added test for inner model hidden
* added autotest decorators
* test mask dtype to torch.int64
* torch.testing.assert_close
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* torch_device and @torch_gpu in tests
* upd tests
* +torch decorators
* torch decorators fixed
* more decorators!
* even more decorators
* fewer decorators
---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Add a convenience method for building in your own name scope
* Second attempt at auto layer building
* Revert "Second attempt at auto layer building"
This reverts commit e03a3aaecf9ec41a805582b83cbdfe3290a631be.
* Attempt #3
* Revert "Attempt #3"
This reverts commit b9df7a0857560d29b5abbed6127d9e9eca77cf47.
* Add missing attributes that we're going to need later
* Add some attributes we're going to need later
* A fourth attempt! Feel the power flow through you!
* Revert "A fourth attempt! Feel the power flow through you!"
This reverts commit 6bf4aaf3875d6f28485f50187617a4c616c8aff7.
* Add more values we'll need later
* TF refactor that we'll need later
* Revert "TF refactor that we'll need later"
This reverts commit ca07202fb5b7b7436b893baa8d688b4f348ea7b9.
* Revert "Revert "TF refactor that we'll need later""
This reverts commit 1beb0f39f293ed9c27594575e1c849aadeb15c13.
* make fixup
* Attempt five!
* Revert "Attempt five!"
This reverts commit 3302207958dfd0374b0447a51c06eea51a506044.
* Attempt six - this time don't add empty methods
* Revert "Attempt six - this time don't add empty methods"
This reverts commit 67d60129be75416b6beb8f47c7d38d77b18d79bb.
* Attempt seven - better base model class detection!
* Revert "Attempt seven - better base model class detection!"
This reverts commit 5f14845e92ea0e87c598da933bfbfee10f553bc9.
* Another attribute we'll need later
* Try again with the missing attribute!
* Revert "Try again with the missing attribute!"
This reverts commit 760c6f30c5dffb3e04b0e73c34a77d1882a0fef7.
* This is the attempt that will pierce the heavens!
* Revert "This is the attempt that will pierce the heavens!"
This reverts commit c868bb657de057aca7a5260350a3f831fc4dfee6.
* Attempt seven - snag list is steadily decreasing
* Revert "Attempt seven - snag list is steadily decreasing"
This reverts commit 46fbd975deda64429bfb3e5fac4fc0370c00d316.
* Attempt eight - will an empty snag list do it?
* Revert "Attempt eight - will an empty snag list do it?"
This reverts commit 7c8a3c2b083253649569e9877e02054ae5cec67b.
* Fixes to Hubert issues that cause problems later
* Trying again with Conv1D/SeparableConv fixes
* Revert "Trying again with Conv1D/SeparableConv fixes"
This reverts commit 55092bca952bc0f750aa1ffe246a640bf1e2036e.
* Apply the build shape fixes to Wav2Vec2 as well
* One more attempt!
* Revert "One more attempt!"
This reverts commit 5ac3e4cb01b9458cc93312873725f9444ae7261c.
* Another attempt!
* Revert "Another attempt!"
This reverts commit ea16d890e019d7de8792a3b8e72f3b1c02adae50.
* Let's see how many failures we get without the internal build method
* Fix OpenAI
* Fix MobileBERT
* (Mostly) fix GroupVIT
* Fix BLIP
* One more BLIP fix
* One more BLIP fix!
* Fix Regnet
* Finally fully fix GroupViT
* Fix Data2Vec and add the new AdaptivePool
* Fix Segformer
* Fix Albert
* Fix Deberta/DebertaV2
* Fix XLM
* Actually fix XLM
* Fix Flaubert
* Fix lxmert
* Fix Resnet
* Fix ConvBERT
* Fix ESM
* Fix Convnext / ConvnextV2
* Fix SAM
* Fix Efficientformer
* Fix LayoutLMv3
* Fix speech_to_text
* Fix mpnet and mobilevit
* Fix Swin
* Fix CTRL
* Fix CVT
* Fix DPR
* Fix Wav2Vec2
* Fix T5
* Fix Hubert
* Fix GPT2
* Fix Whisper
* Fix DeiT
* Fix the encoder-decoder / dual-encoder classes
* make fix-copies
* build in name scope
* Fix summarization test
* Fix tied weight names for BART + Blenderbot
* Fix tied weight name building
* Fix to TFESM weight building
* Update TF SAM
* Expand all the shapes out into Big Boy Shapes
* fix a typo and add an illustrative test
* appease black
* reduce code duplication and add Annotion type back with a pending deprecation warning
* remove unused code
* change warning type
* black formatting fix
* change enum deprecation approach to support 3.8 and earlier
* add stacklevel
* fix black issue
* fix ruff issues
* fix ruff issues
* move tests to own mixin
* include yolos
* fix black formatting issue
* fix black formatting issue
* use logger instead of warnings and include target version for deprecation
* Skip nn.Module.reset_parameters
* Actually skip
* Check quality
* Maybe change all inits
* Fix init issues: only modify public functions
* Add a small test for now
* Style
* test updates
* style
* nice tes
* style
* make it even faster
* one more second
* remove fx icompatible
* Update tests/test_modeling_common.py
Co-authored-by: Lysandre Debut <hi@lysand.re>
* Update tests/test_modeling_common.py
Co-authored-by: Lysandre Debut <hi@lysand.re>
* skip
* fix quality
* protect the import
---------
Co-authored-by: Lysandre Debut <hi@lysand.re>
* add sdpa
* wip
* cleaning
* add ref
* yet more cleaning
* and more :)
* wip llama
* working llama
* add output_attentions=True support
* bigcode sdpa support
* fixes
* gpt-bigcode support, require torch>=2.1.1
* add falcon support
* fix conflicts falcon
* style
* fix attention_mask definition
* remove output_attentions from attnmaskconverter
* support whisper without removing any Copied from statement
* fix mbart default to eager renaming
* fix typo in falcon
* fix is_causal in SDPA
* check is_flash_attn_2_available in the models init as well in case the model is not initialized through from_pretrained
* add warnings when falling back on the manual implementation
* precise doc
* wip replace _flash_attn_enabled by config.attn_implementation
* fix typo
* add tests
* style
* add a copy.deepcopy on the config in from_pretrained, as we do not want to modify it inplace
* obey to config.attn_implementation if a config is passed in from_pretrained
* fix is_torch_sdpa_available when torch is not installed
* remove dead code
* Update src/transformers/modeling_attn_mask_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_attn_mask_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_attn_mask_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_attn_mask_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_attn_mask_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/bart/modeling_bart.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* remove duplicate pretraining_tp code
* add dropout in llama
* precise comment on attn_mask
* add fmt: off for _unmask_unattended docstring
* precise num_masks comment
* nuke pretraining_tp in LlamaSDPAAttention following Arthur's suggestion
* cleanup modeling_utils
* backward compatibility
* fix style as requested
* style
* improve documentation
* test pass
* style
* add _unmask_unattended tests
* skip meaningless tests for idefics
* hard_check SDPA requirements when specifically requested
* standardize the use if XXX_ATTENTION_CLASSES
* fix SDPA bug with mem-efficient backend on CUDA when using fp32
* fix test
* rely on SDPA is_causal parameter to handle the causal mask in some cases
* fix FALCON_ATTENTION_CLASSES
* remove _flash_attn_2_enabled occurences
* fix test
* add OPT to the list of supported flash models
* improve test
* properly test on different SDPA backends, on different dtypes & properly handle separately the pad tokens in the test
* remove remaining _flash_attn_2_enabled occurence
* Update src/transformers/modeling_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/modeling_attn_mask_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update docs/source/en/perf_infer_gpu_one.md
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* remove use_attn_implementation
* fix docstring & slight bug
* make attn_implementation internal (_attn_implementation)
* typos
* fix tests
* deprecate use_flash_attention_2=True
* fix test
* add back llama that was removed by mistake
* fix tests
* remove _flash_attn_2_enabled occurences bis
* add check & test that passed attn_implementation is valid
* fix falcon torchscript export
* fix device of mask in tests
* add tip about torch.jit.trace and move bt doc below sdpa
* fix parameterized.expand order
* move tests from test_modeling_attn_mask_utils to test_modeling_utils as a relevant test class is already there
* update sdpaattention class with the new cache
* Update src/transformers/configuration_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/bark/modeling_bark.py
* address review comments
* WIP torch.jit.trace fix. left: test both eager & sdpa
* add test for torch.jit.trace for both eager/sdpa
* fix falcon with torch==2.0 that needs to use sdpa
* fix doc
* hopefully last fix
* fix key_value_length that has no default now in mask converter
* is it flacky?
* fix speculative decoding bug
* tests do pass
* fix following #27907
---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Fix issues in add and is_done for BeamHypotheses
* make newly added arguments optional for better compatibility
* Directly use cur_len as generated_len, add note for retrocompatibility
* update test expectation
* make cur_len represents the length of the entire sequence including the decoder prompt
* remove redundant if/else in testing
* Draft version of new KV Caching
This should allow Attention Sinks (https://github.com/tomaarsen/attention_sinks)
/ StreamingLLM (https://arxiv.org/abs/2309.17453) to be easily implemented
in a third-party or in transformers directly
* Address numerous PR suggestions
1. Move layer_idx from cache to ...Attention. Removes confusing set_layer_idx magic.
2. Always convert past_key_values to Cache instance at the start of ...Attention, removes all other isinstance calls.
3. Remove __bool__ and __getitem__ magic as they're confusing.
4. past_key_values.update(key, value, idx) now returns key, value.
5. Add use_legacy_cache flag, defaults to None, i.e. Falsey. This breaks generate for now, until 1) the cache is used is generate() or 2) use_legacy_cache is defaulted to True in generate() until we change it in another PR.
6. Separate key_cache and value_cache.
Some work is still needed to see if the SinkCache can conveniently be implemented with just one update method.
* Implement the SinkCache through backward+forward rotations
* Integrate (Sink)Cache with Llama FA2
* Set use_legacy_cache=True as default, allows for test passes
* Move from/to_legacy_cache to ...Model class
* Undo unnecessary newline change
* Remove copy utility from deprecated OpenLlama
* Match import style
* manual rebase with main
* Cache class working with generate (#1)
* Draft version of new KV Caching
This should allow Attention Sinks (https://github.com/tomaarsen/attention_sinks)
/ StreamingLLM (https://arxiv.org/abs/2309.17453) to be easily implemented
in a third-party or in transformers directly
* Address numerous PR suggestions
1. Move layer_idx from cache to ...Attention. Removes confusing set_layer_idx magic.
2. Always convert past_key_values to Cache instance at the start of ...Attention, removes all other isinstance calls.
3. Remove __bool__ and __getitem__ magic as they're confusing.
4. past_key_values.update(key, value, idx) now returns key, value.
5. Add use_legacy_cache flag, defaults to None, i.e. Falsey. This breaks generate for now, until 1) the cache is used is generate() or 2) use_legacy_cache is defaulted to True in generate() until we change it in another PR.
6. Separate key_cache and value_cache.
Some work is still needed to see if the SinkCache can conveniently be implemented with just one update method.
* Integrate (Sink)Cache with Llama FA2
* Move from/to_legacy_cache to ...Model class
* Undo unnecessary newline change
* Match import style
* working generate
* Add tests; Simplify code; Apply changes to Mistral and Persimmon
* fix rebase mess
* a few more manual fixes
* last manual fix
* propagate changes to phi
* upgrade test
* add use_legacy_cache docstring; beef up tests
* reintroduce unwanted deletes
---------
Co-authored-by: Tom Aarsen <Cubiegamedev@gmail.com>
* move import
* add default to model_kwargs.get('use_legacy_cache')
* correct failing test
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* apply PR suggestions
* fix failing test
* Apply suggestions from code review
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: Tom Aarsen <37621491+tomaarsen@users.noreply.github.com>
* PR comments
* tmp commit
* add docstrings
* more tests, more docstrings, add to docs
* derp
* tmp commit
* tmp dbg
* more dbg
* fix beam search bug
* cache can be a list of tuples in some models
* fix group beam search
* all but sinkcache integration tests
* fix sink cache and add hard integration test
* now also compatible with input_embeds input
* PR comments
* add Cache support to Phi+FA2
* make fixup
---------
Co-authored-by: Joao Gante <joao@huggingface.co>
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
