* Fix llama model forward function with attention=True, same-length encoded sequence.
* Fix style
* propagate fix to modeling_cohere, gemma, dbrx, and olmo (which copy the same sdpa masking logic from llama)
* Fix style
* ignore unnecessary sdpa mask converter when output_attentions=True
* add tests checking sdpa and eager outputs match when output_attentions=True
* Split if statements in two lines
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Fix formatting
* Add fix to new jetmoe model
* Add missing output_attentions argument to jetmoe mask creation
---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Adding _tie_weights() to prediction heads to support low_cpu_mem_usage=True
* Testing for the non-safe-tensors case, since the default is safe-tensors already
* Running fixup/fix-copies
* Adding accelerate annotations to tests
* move scaling to nn.Module
* let the test be here for now (need to fix)
* failing tests
* last failing models
* Revert commit 4c14817f38
* clean-up
* oops forgot
* codestyle
* raise NotImplemented when possible
* Update tests/test_modeling_common.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* skip tests in respective modeling files
---------
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* Adding SDPA support for BERT
* Using the proper input name for testing model input in inference()
* Adding documentation for SDPA in BERT model page
* Use the stable link for the documentation
* Adding a gate to only call .contiguous() for torch < 2.2.0
* Additions and fixes to the documentation
* Minor updates to documentation
* Adding extra requirements needed for the contiguous() bug
* Adding "Adapted from" in plcae of the "Copied from"
* Add benchmark speedup tables to the documentation
* Minor fixes to the documentation
* Use ClapText as a replacemenet for Bert in the Copied-From
* Some more fixes for the fix-copies references
* Overriding the test_eager_matches_sdpa_generate in bert tests to not load with low_cpu_mem_usage
[test all]
* Undo changes to separate test
* Refactored SDPA self attention code for KV projections
* Change use_sdpa to attn_implementation
* Fix test_sdpa_can_dispatch_on_flash by preparing input (required for MultipleChoice models)
* feat: multidevice for resnet
* feat: yes! resnet
* fix: compare all elements in tuple
* feat: support for regnet
* feat: support for convnextv2
* feat: support for bit
* feat: support for cvt
* feat: add support for focalnet
* feat: support for yolos
* feat: support for glpn
* feat: support for imagegpt
* feat: support for levit
* feat: support for mgp_str
* feat: support for mobilnet_v1
* feat: support for mobilnet_v2
* feat: support for mobilevit
* feat: support for mobilevitv2
* feat: support for poolformer
* fix: copies
* fix: code quality check
* update: upstream changes from main
* fix: consistency check
* feat: support for sam
* feat: support for switchformer
* feat: support for swin
* feat: support for swinv2
* feat: support for timesformer
* feat: suport for trocr
* feat: support for upernet
* fix: check copies
* update: rerun CI
* update: rerun again, maybe
* update: one more rerun
---------
Co-authored-by: Jacky Lee <jackylee328@gmail.com>
* add tests for batching support
* Update src/transformers/models/fastspeech2_conformer/modeling_fastspeech2_conformer.py
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
* Update src/transformers/models/fastspeech2_conformer/modeling_fastspeech2_conformer.py
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
* Update tests/test_modeling_common.py
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
* Update tests/test_modeling_common.py
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
* Update tests/test_modeling_common.py
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
* fixes and comments
* use cosine distance for conv models
* skip mra model testing
* Update tests/models/vilt/test_modeling_vilt.py
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
* finzalize and make style
* check model type by input names
* Update tests/models/vilt/test_modeling_vilt.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* fixed batch size for all testers
* Revert "fixed batch size for all testers"
This reverts commit 525f3a0a05.
* add batch_size for all testers
* dict from model output
* do not skip layoutlm
* bring back some code from git revert
* Update tests/test_modeling_common.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* Update tests/test_modeling_common.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* clean-up
* where did minus go in tolerance
* make whisper happy
* deal with consequences of losing minus
* deal with consequences of losing minus
* maskformer needs its own test for happiness
* fix more models
* tag flaky CV models from Amy's approval
* make codestyle
---------
Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* Revert "Add tie_weights() to LM heads and set bias in set_output_embeddings() (#28948)"
This reverts commit 725f4ad1cc.
* Revert "Patch to skip failing `test_save_load_low_cpu_mem_usage` tests (#29043)"
This reverts commit 4156f517ce.
* Add tie_weights() to LM heads and set bias in set_output_embeddings()
The bias were not tied correctly in some LM heads, and this change should fix that.
* Moving test_save_and_load_low_cpu_mem_usage to ModelTesterMixin
* Adding _tie_weights() to MPNet and Vilt
* Skip test for low cpu mem usage for Deta/DeformableDetr since they cannot init on meta device
* Rename to test name to save_load to match the convention
* test that tied output embeddings aren't initialized on load
* don't initialize the output embeddings if we're going to tie them to the input embeddings
* skip bf16 test if not supported by device
* fix
* fix bis
* use is_torch_bf16_available_on_device
* use is_torch_fp16_available_on_device
* fix & use public llama
* use 1b model
* fix flacky test
---------
Co-authored-by: Your Name <you@example.com>
* 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>
* 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>
