* 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>
* Add X-MOD to Readme
* Add documentation for X-MOD
* Implement X-MOD
* Fix formatting of X-MOD docs
* Change signature of X-MOD forward methods to use lang_ids
* Minor changes
* Rebase with main and run make fix-copies
* Make suggested changes to docstrings
* Improve code readability
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
* Fix code style
* Conversion script: Remove asserts and type annotations
* Remove _TOKENIZER_FOR_DOC
* XMOD -> Xmod
* Update copyright note
* Fix doctests
* Fix docstring
* Add integration test for FillMaskPipeline
* Revert "Add integration test for FillMaskPipeline"
This reverts commit 4381eb3b1d0f5d85785f89caba83928e6efa6d1f.
* Add end-to-end integration test for mask fill
* make style
* Rebase with main and make fix-copies
---------
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
* Copy RoBERTa
* formatting
* implement RoBERTa with prelayer normalization
* update test expectations
* add documentation
* add convertion script for DinkyTrain weights
* update checkpoint repo
Unfortunately the original checkpoints assumes a hacked roberta model
* add to RoBERTa-PreLayerNorm docs to toc
* run utils/check_copies.py
* lint files
* remove unused import
* fix check_repo reporting wrongly a test is missing
* fix import error, caused by rebase
* run make fix-copies
* add RobertaPreLayerNormConfig to ROBERTA_EMBEDDING_ADJUSMENT_CONFIGS
* Fix documentation <Facebook> -> Facebook
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* fixup: Fix documentation <Facebook> -> Facebook
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Add missing Flax header
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* expected_slice -> EXPECTED_SLICE
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* update copies after rebase
* add missing copied from statements
* make fix-copies
* make prelayernorm explicit in code
* fix checkpoint path for the original implementation
* add flax integration tests
* improve docs
* update utils/documentation_tests.txt
* lint files
* Remove Copyright notice
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* make fix-copies
* Remove EXPECTED_SLICE calculation comments
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Add templates for gpt-sw3
* Add templates for gpt-sw3
* Added sentencepiece tokenizer
* intermediate commit with many changes
* fixed conflicts
* Init commit for tokenization port
* Tokenization progress
* Remove fast tokenizer
* Clean up and rename spm.model -> spiece.model
* Remove TF -> PT conversion script template, Clean up Megatron -> PT script
* Optimize encode & decode performance
* added new attention
* added new attention
* attention for gpt-sw3 working
* attention good
* Cache is now working
* fixed attention mask so that it works with causal attention
* fixed badbmm bug for cpu and caching
* updated config with correct parameters
* Refactor and leave optimizations as separate functions to avoid breaking expected functionality
* Fix special tokens mapping for both tokenizers
* cleaning up of code and comments
* HF compatible attention outputs
* Tokenizer now passing tests, add documentation
* Update documentation
* reverted back to base implementation after checking that it is identical to pretrained model
* updated gpt-sw3 config
* updated conversion script
* aligned parameters with gpt-sw3 config
* changed default scale_attn_by_inverse_layer_idx to true
* removed flag from conversion script
* added temporary model path
* reverted back to functioning convert script
* small changes to default config
* updated tests for gpt-sw3
* make style, make quality, minor cleanup
* Change local paths to testing online repository
* Change name: GptSw3 -> GPTSw3
* Remove GPTSw3TokenizerFast references
* Use official model repository and add more model sizes
* Added reference to 6.7b model
* Add GPTSw3DoubleHeadsModel to IGNORE_NON_AUTO_CONFIGURED, like GPT2DoubleHeadsModel
* Remove pointers to non-existing TFGPTSw3
* Add GPTSw3 to docs/_toctree.yml
* Remove TF artifacts from GPTSw3 in __init__ files
* Update README:s with 'make fix-copies'
* Add 20b model to archive list
* Add documentation for GPT-Sw3
* Fix typo in documentation for GPT-Sw3
* Do 'make fix-copies' again after having updated docs
* Fix some typos in docs
* Update src/transformers/models/gpt_sw3/configuration_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/configuration_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/__init__.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/__init__.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/convert_megatron_to_pytorch.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/modeling_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update tests/models/gpt_sw3/test_tokenization_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/modeling_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/modeling_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Resolve comments from PR feedback
* Resolve more comments from PR feedback, also set use_cache=True in convert script
* Add '# Copied from' comments for GPTSw3 modeling
* Set 'is_parallelizable = False'
* Remove '# Copied from' where code was modified and add 'with x->y' when appropriate
* Remove parallelize in mdx
* make style, make quality
* Update GPTSw3Config default values and corresponding documentation
* Update src/transformers/models/gpt_sw3/tokenization_gpt_sw3.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/__init__.py
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Clean up and protect GPTSw3Tokenizer imports with is_sentencepiece_available
* Make style, make quality
* Add dummy object for GPTSw3Tokenizer via 'make fix-copies'
* make fix-copies
* Remove GPTSw3 modeling classes
* make style, make quality
* Add GPTSw3 auto-mappings for other GPT2 heads
* Update docs/source/en/model_doc/gpt-sw3.mdx
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/convert_megatron_to_pytorch.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/models/gpt_sw3/tokenization_gpt_sw3.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Remove old TODO-comment
* Add example usage to GPTSw3Tokenizer docstring
* make style, make quality
* Add implementation details and example usage to gpt-sw3.mdx
Co-authored-by: JoeyOhman <joeyoh@kth.se>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* add model files etc for MobileNetV2
rename files for MobileNetV1
initial implementation of MobileNetV1
fix conversion script
cleanup
write docs
tweaks
fix conversion script
extract hidden states
fix test cases
make fixup
fixup it all
remove main from doc link
fixes
fix tests
fix up
use google org
fix weird assert
* fixup
* use google organization for checkpoints
* add model files etc for MobileNetV2
* rename files for MobileNetV1
* initial implementation of MobileNetV1
* fix conversion script
* cleanup
* write docs
* tweaks
* fix conversion script
* extract hidden states
* fix test cases
* make fixup
* fixup it all
* rename V1 to V2
* fix checkpoints
* fixup
* implement first block + weight conversion
* add remaining layers
* add output stride and dilation
* fixup
* add tests
* add deeplabv3+ head
* a bit of fixup
* finish deeplab conversion
* add link to doc
* fix issue with JIT trace
in_height and in_width would be Tensor objects during JIT trace, which caused Core ML conversion to fail on the remainder op. By making them ints, the result of the padding calculation becomes a constant value.
* cleanup
* fix order of models
* fix rebase error
* remove main from doc link
* add image processor
* remove old feature extractor
* fix converter + other issues
* fixup
* fix unit test
* add to onnx tests (but these appear broken now)
* add post_process_semantic_segmentation
* use google org
* remove unused imports
* move args
* replace weird assert
* docs: Fix typo in ONNX parser help: 'tolerence' => 'tolerance'
* docs: Resolve many typos in the English docs
Typos found via 'codespell ./docs/source/en'
* Implement ONNX support for Longformer
Fix repo consistency check complaints
Fix value mismatches
Add pooler output for default model
Increase validation atol to accommodate multiple-choice error
Fix copies
Fix chunking for longer sequence lengths
Add future comment
* Fix issue in mask_invalid_locations
* Remove torch imports in configuration_longformer
* Change config access to fix LED
* Push opset version to support tril
* Work in review comments (mostly style)
* Add Longformer to ONNX tests
* onnx config for clip
* default opset as 14
* changes from the original repo
* input values order fix
* outputs fix
* remove unused import
* ran make fix-copies
* black format
* review comments: forward ref, import fix, model change revert, .to cleanup
* make style
* formatting fixes
* revert groupvit
* comment for cast to int32
* comment fix
* make .T as .t() for onnx conversion
* ran make fix-copies
* remove unneeded comment
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* fix copies
* remove comment
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* update features
* MT5OnnxConfig added with updated with tests and docs
* fix imports
* fix onnc_config_cls for mt5
Co-authored-by: Thomas Chaigneau <thomas.deeptools.ai>
* add onnx support for BLOOM
* use TYPE_CHECKING for type annotations
* fix past_shape for bloom (different from gpt2)
* use logical_or instead of `+` for onnx support
* bigger `atol_for_validation` for larger bloom models
* copied -> taken because it's no longer an exact copy
* remove "copied from" comment
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
* Add CodeGen model
* Add missing key and switch order of super()
* Fix torch.ones init with uint8 instead of bool
* Address comments: copy statements and doc
* update tests
* remove old model parallel
* fix batch gen tests
* fix batch gen test
* update test_gpt2_sample_max_time
* fix codgen test and revert gpt2 test change
* Fix incorrect tie_word_embedding value, typo, URL
* Fix model order in README and styling
* Reorder model list alphabetically
* Set tie_word_embedding to False by default
* Apply suggestions from code review
* Better attn mask name & remove attn masked_bias
* add tokenizer for codegen
* quality
* doc tokenizer
* fix-copies
* add CodeGenTokenizer in converter
* make truncation optional
* add test for truncation
* add copyright
* fix-copies
* fix fast tokenizer decode
* Update src/transformers/models/codegen/tokenization_codegen.py
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* increase vocab_size in tests
Co-authored-by: patil-suraj <surajp815@gmail.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* Initial commit
* Make some fixes
* Make PT model full forward pass
* Drop TF & Flax implementation, fix copies etc
* Add Flax model and update some corresponding stuff
* Drop some TF things
* Update config and flax local attn
* Add encoder_attention_type to config
* .
* Update docs
* Do some cleansing
* Fix some issues -> make style; add some docs
* Fix position_bias + mask addition + Update tests
* Fix repo consistency
* Fix model consistency by removing flax operation over attn_mask
* [WIP] Add PT TGlobal LongT5
* .
* [WIP] Add flax tglobal model
* [WIP] Update flax model to use the right attention type in the encoder
* Fix flax tglobal model forward pass
* Make the use of global_relative_attention_bias
* Add test suites for TGlobal model
* Fix minor bugs, clean code
* Fix pt-flax equivalence though not convinced with correctness
* Fix LocalAttn implementation to match the original impl. + update READMEs
* Few updates
* Update: [Flax] improve large model init and loading #16148
* Add ckpt conversion script accoring to #16853 + handle torch device placement
* Minor updates to conversion script.
* Typo: AutoModelForSeq2SeqLM -> FlaxAutoModelForSeq2SeqLM
* gpu support + dtype fix
* Apply some suggestions from code review
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
* * Remove (de)parallelize stuff
* Edit shape comments
* Update README.md
* make fix-copies
* Remove caching logic for local & tglobal attention
* Apply another batch of suggestions from code review
* Add missing checkpoints
* Format converting scripts
* Drop (de)parallelize links from longT5 mdx
* Fix converting script + revert config file change
* Revert "Remove caching logic for local & tglobal attention"
This reverts commit 2a619828f6ddc3e65bd9bb1725a12b77fa883a46.
* Stash caching logic in Flax model
* Make side relative bias used always
* Drop caching logic in PT model
* Return side bias as it was
* Drop all remaining model parallel logic
* Remove clamp statements
* Move test files to the proper place
* Update docs with new version of hf-doc-builder
* Fix test imports
* Make some minor improvements
* Add missing checkpoints to docs
* Make TGlobal model compatible with torch.onnx.export
* Replace some np.ndarray with jnp.ndarray
* Fix TGlobal for ONNX conversion + update docs
* fix _make_global_fixed_block_ids and masked neg value
* update flax model
* style and quality
* fix imports
* remove load_tf_weights_in_longt5 from init and fix copies
* add slow test for TGlobal model
* typo fix
* Drop obsolete is_parallelizable and one warning
* Update __init__ files to fix repo-consistency
* fix pipeline test
* Fix some device placements
* [wip]: Update tests -- need to generate summaries to update expected_summary
* Fix quality
* Update LongT5 model card
* Update (slow) summarization tests
* make style
* rename checkpoitns
* finish
* fix flax tests
Co-authored-by: phungvanduy <pvduy23@gmail.com>
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>
Co-authored-by: Patrick von Platen <patrick.v.platen@gmail.com>
Co-authored-by: patil-suraj <surajp815@gmail.com>
* Add onnx configuration for xlm
* Add supported features for xlm
* Add xlm to models exportable with onnx
* Add xlm architecture to test file
* Modify docs
* Make code quality fixes
* Automatically sort auto mappings
* Better class extraction
* Some auto class magic
* Adapt test and underlying behavior
* Remove re-used config
* Quality
