* add draft changes
* fix failing wav2vec
* style
* make sure that the argument is saved + add tests
* style
* fixup
* update test
* default clean_up_tokenization_spaces to False for Bloom and Llama
* Update code based on review
Co-authored-by: Nicolas Patry <patry.nicolas@gmail.com>
* style
* quality
---------
Co-authored-by: Nicolas Patry <patry.nicolas@gmail.com>
* ensure causal_mask is created directly on device
* add copy tag to opt, update bart implementation
* add device to all _make_causal_mask copies
* formatting fixes
* more manual fixes due to unlinked versions of _prepare_decoder_attention_mask
* [neptune] fix checkpoint bug with relative out_dir
* update imports
* reformat with black
* check neptune without imports
* fix typing-related issue
* run black on code
* use os.path.sep instead of raw \
* simplify imports and remove type annotation
* make ruff happy
* apply review suggestions
---------
Co-authored-by: Aleksander Wojnarowicz <alwojnarowicz@gmail.com>
* Initial commit
* update modeling code
* update doc
* add functions necessary
* fix impotrs
* revert changes
* fixup
* more styling to get going
* remove standalone encoder
* update code
* styling
* fix config and model
* update code and some refactoring
* make more tests pass
* Adding NLLB-200 - MoE - 54.5B for no language left behind
Fixes#21300
* fix mor common tests
* styke
* update testing file
* update
* update
* Router2 doc
* update check config with sparse layer
* add dummy router
* update current conversion script
* create on the fly conversion script
* Fixup
* style
* style 2
* fix empty return
* fix return
* Update default config sparse layers
* easier to create sparse layers
* update
* update conversion script
* update modeling
* add to toctree
* styling
* make ruff happy
* update docstring
* update conversion script
* update, will break tests but impelemting top2
* update
* ❗local groups are supported here
* ⚠️ Support for local groups is now removed ⚠️
This is because it has to work with model parallelism that we do not support
* finish simplificaiton
* Fix forward
* style
* fixup
* Update modelling and test, refactoring
* update tests
* remove final layer)norm as it is done in the FF
* routing works! Logits test added
* nit in test
* remove top1router
* style
* make sure sparse are tested. Had to change route_tokens a liottle bit
* add support for unslip models when converting
* fixup
* style
* update test s
* update test
* REFACTOR
* encoder outputs match!
* style
* update testing
* 🎉encoder and decoder logits match 🎉
* styleing
* update tests
* cleanup tests
* fix router test and CIs
* cleanup
* cleanup test styling
* fix tests
* Finally the generation tests match!
* cleanup
* update test
* style testing file
* remove script
* cleanup
* more cleanup
* nits
* update
* NLLB tokenizer is wrong and will be fixed soon
* use LongTensors
* update tests
* revert some small changes
* fix second expert sampling and batch prioritized routing
* update tests
* finish last tests
* make ruff happy
* update
* ruff again
* style
* Update docs/source/en/model_doc/nllb-moe.mdx
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Updates based on review
* style and fix import issue
* nit
* more nits
* cleanup
* styling
* update test_seconde_expert_policy
* fix name
* last nit on the markdown examples
---------
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Add initial remote hardware auto-setup docs
* Fix a few typos and clarify some language
* Add missing dependency
* Update self-hosted launch script with Sylvain's comments.
* Formatting.
* Trigger CI
* Style
* First draft
* Fix integration test
* Remove script
* Fix test and typos
* Fix one more test
* Skip tied embeddings test
* Remove line
* Address comments
* Wav2Vec2ProcessorWithLM can return N best hypotheses now
Signed-off-by: Vladislav Sokolovskii <vladislav@parrothq.com>
* Wav2Vec2ProcessorWithLM n_best cannot be None
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
* Batch decoding can return N best hypotheses now
batch_decode was extended with the same functionality as decode
function, N best hypotheses per sample can be returned
Signed-off-by: Vladislav Sokolovskii <vladislav@parrothq.com>
---------
Signed-off-by: Vladislav Sokolovskii <vladislav@parrothq.com>
Co-authored-by: Vladislav Sokolovskii <vladislav@parrothq.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>
Edited one line in src/transormers/generation/utils.py. Changed dist.world_size() to dist.get_world_size() since world_size() doesn't exist in pytorch.dist.
* add mega file structure and plain pytorch version of mega source code
* added config class with old naming conventions
* filled in mega documentation
* added config class and embeddings with optional token types
* updated notes
* starting the conversion process, deleted intermediate and added use_cache back to config
* renamed config attributes in modeling_mega.py
* checkpointing before refactoring incremental decoding functions
* removed stateful incremental key/values for EMA and self-attention
* refactored MovingAverageGatedAttention to remove stateful k/v history and use unified attention mask
* MovingAverageGatedAttention works with incremental decoding + past values, added sequence length enforcement
* more comments in MovingAverageGatedAttention + checkpointing before GatedCrossAttention
* bug fix in attention mask handling in MovingAverageGatedAttention
* removed incremental state from GatedCrossAttention and removed IncrementalState class
* finished gated cross attention and got MegaLayer working
* fixed causal masking in mega decoder
* fixed how padding and causal masks are passed through MegaLayer with and without k/v caching
* finished MegaModel; tested with encoder, decoder-only, and cross-attention type inputs; started work on downstream classes; removed mentions of position_ids
* added optional dense hidden layer for masked and causal LM classes
* docstring updates in MultiHeadEMA and GatedCrossAttention, removed unnecessary inputs in cross-attention
* removed before_attn_fn in Mega class and updated docstrings and comments up to there
* bug fix in MovingAverageGatedAttention masking
* working conversion of MLM checkpoint in scratchpad script -- perfect matches
* moved arg for hidden dense layer in LM head to config; discovered issue where from_pretrained is renaming gamma and beta parameters
* renamed gamma and beta parameters to avoid HF renaming when loading from checkpoint
* finished checkpoint conversion script
* cleanup old class in mega config script
* removed 'copied from' statements and passing integration tests
* added num_attention_heads=1 to config for integration compatibility, decoder tests working, generation tests failing
* fixed tuple output of megamodel
* all common tests passing after fixing issues in decoder, gradient retention, and initialization
* added mega-specific tests, ready for more documentation and style checks
* updated docstrings; checkpoint before style fixes
* style and quality checks, fixed initialization problem in float_tensor, ready for PR
* added mega to toctree
* removed unnecessary arg in megaconfig
* removed unused arg and fixed code samples with leftover roberta models
* Apply suggestions from code review
Applied all suggestions except the one renaming a class, as I'll need to update that througout
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* fixed issue where .view breaks batch dimension, conversion script fixed with absolute imports, updated readme with Mega->MEGA
* removed asserts in Mega code, renamed sequencenorm, gatedcrossattention, and NFFN, replaced get_activation_fn with ACTFN, and added sequencenorm to layer norms
* reformatted .forward() docstrings to match style and removed unused mask input in cross-attention
* removed all reset_parameters() methods and rolled into MegaPreTrainedModel._init_weights()
* renamed all single-letter variables and improved readability in tensor size comments, Mega->MEGA in 2 documentation files
* variable names in NFFN
* manual Mega->MEGA changes in docs
* Mega->MEGA in config auto
* style and quality fixes
* Apply suggestions from code review
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* renamed parameters and variables with confusing names, added copied from statements, moved fft conv to its own method, other cleanup from PR comments
* commit before dealing with merge conflicts
* made new attention activation functions available in ACT2FN and added generation test from OPT
* style and quality in activations and tests
* documentation fixes, renaming variables in dropout and rotary positions, used built-in causal masking, encoders->layers in MegaModel, moved comments into docstrings
* style and quality fixes after latest updates, before rotary position ids
* causal mask in MegaBlock docstring + added missing device passing
* Apply suggestions from code review
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update README.md
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* added Mega prefixes where missing, reverted MegaSequenceNorm to if-else, other module renaming requested in PR
* style and quality fixes + readme updates pointing to main
---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
