GitLab

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

* Use Llama RoPE implementation for Falcon

+ Add copy functionalities

* Use standard cache format for Falcon

* Simplify apply_rotary_pos_emb, copy from Llama

* Remove unnecessary cache conversion test

We don't need to convert any caches anymore!

* Resolve copy complaint

* Revert "Falcon: fix revision propagation (#26006)"

This reverts commit 118c676ef3124423e5d062b665f05cde55bc9a90.

* Revert "Put Falcon back (#25960)"

This reverts commit 22a69f1d.

* Put Falcon back

* Update src/transformers/models/auto/configuration_auto.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update test

---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* CI with layers=2

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

* update tiny_model_summary.json

* update

* update

* update

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

* Initial commit

* Update src/transformers/models/falcon/configuration_falcon.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Update src/transformers/models/falcon/configuration_falcon.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Cleanup config docstring

* Update src/transformers/models/falcon/configuration_falcon.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* Convert to relative imports

* Remove torch < 1.8 warning

* Restructure cos_sin header

* qkv -> query, key, value

* Refactor attention calculation

* Add a couple of config variables to account for the different checkpoints

* Successful merging of the code paths!

* Fix misplaced line in the non-parallel attention path

* Update config and tests

* Add a pad_token_id when testing

* Support output_attentions when alibi is None

* make fixup

* Skip KV cache shape test

* No more _keys_to_ignore_on_load_missing

* Simplify self attention a bit

* Simplify self attention a bit

* make fixup

* stash commit

* Some more attention mask updates

* Should pass all tests except assisted generation!

* Add big model generation test

* make fixup

* Add temporary workaround for test

* Test overrides for assisted generation

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Update tests/models/falcon/test_modeling_falcon.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Test overrides for assisted generation

* Add generation demo

* Update copyright

* Make the docstring model actually small

* Add module-level docstring

* Remove all assertions

* Add copied from bloom

* Reformat the QKV layer

* Add copied from bloom

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* Remove unused line and reformat

* No single letter variables

* Cleanup return names

* Add copied from line

* Remove the deprecated arguments blocks

* Change the embeddings test to an alibi on/off test

* Remove position_ids from FalconForQA

* Remove old check for token type IDs

* Fix the alibi path when multi_query is False

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update src/transformers/models/falcon/modeling_falcon.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update tests/models/falcon/test_modeling_falcon.py
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* Update config naming

* Fix typo for new_decoder_architecture

* Add some comments

* Fix docstring

* Fix docstring

* Create range in the right dtype from the start

* Review comment cleanup

* n_head_kv -> num_kv_heads

* self.alibi -> self.use_alibi

* self.num_kv -> self.num_kv_heads

* Reorder config args

* Made alibi arguments Optional

* Add all model docstrings

* Add extra checkpoints

* Add author info for Falcon

* Stop removing token_type_ids because our checkpoints shouldn't return it anymore

* Add one hopeful comment for the future

* Fix typo

* Update tests, fix cache issue for generation

* Use -1e9 instead of -inf to avoid float overflow

* Recompute the rotary embeddings much less often

* Re-enable disabled tests

* One final fix to attention mask calculation, and update tests

* Cleanup targeting falcon-40b equivalency

* Post-rebase docs update

* Update docstrings, especially in the config

* More descriptive variable names, and comments where we can't rename them

---------
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>