* remove it from all py files
* remove it from the doc
* remove it from examples
* style
* remove traces of _fast_init
* Update test_peft_integration.py
* CIs
* no image
* test
* revert jax version updates
* make fixup
* update autodoc path for model_addition_debugger
* shieldgemma2
* add missing pages to toctree
* draft of model tracer visualiser
* add context manager in addition to decorator
* add debug utils to init
* move model debugging utils to dedicated file
* add documentation
* protect some imports
* format
* move and protect imports
* format
* doc: improve errors in case of broken dummy imports.
* format
* use automatic torch backend
* update doc
* fix backend
* (TEMP) move to dummies while backend wait
* update documentation
* doc
* Add SynthIDTextWatermarkLogitsProcessor
* esolving comments.
* Resolving comments.
* esolving commits,
* Improving SynthIDWatermark tests.
* switch to PT version
* detector as pretrained model + style
* update training + style
* rebase
* Update logits_process.py
* Improving SynthIDWatermark tests.
* Shift detector training to wikitext negatives and stabilize with lower learning rate.
* Clean up.
* in for 7B
* cleanup
* upport python 3.8.
* README and final cleanup.
* HF Hub upload and initiaze.
* Update requirements for synthid_text.
* Adding SynthIDTextWatermarkDetector.
* Detector testing.
* Documentation changes.
* Copyrights fix.
* Fix detector api.
* ironing out errors
* ironing out errors
* training checks
* make fixup and make fix-copies
* docstrings and add to docs
* copyright
* BC
* test docstrings
* move import
* protect type hints
* top level imports
* watermarking example
* direct imports
* tpr fpr meaning
* process_kwargs
* SynthIDTextWatermarkingConfig docstring
* assert -> exception
* example updates
* no immutable dict (cant be serialized)
* pack fn
* einsum equivalent
* import order
* fix test on gpu
* add detector example
---------
Co-authored-by: Sumedh Ghaisas <sumedhg@google.com>
Co-authored-by: Marc Sun <marc@huggingface.co>
Co-authored-by: sumedhghaisas2 <138781311+sumedhghaisas2@users.noreply.github.com>
Co-authored-by: raushan <raushan@huggingface.co>
* clean-up
* Update src/transformers/cache_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/cache_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* Update src/transformers/cache_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* fixup
* Update tests/quantization/quanto_integration/test_quanto.py
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
* Update src/transformers/generation/configuration_utils.py
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
* more suggestions
* mapping if torch available
* run tests & add 'support_quantized' flag
* fix jamba test
* revert, will be fixed by another PR
* codestyle
* HQQ and versatile cache classes
* final update
* typo
* make tests happy
---------
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>
Co-authored-by: Younes Belkada <49240599+younesbelkada@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>
* added informer to gitignore
* added informer to gitignore
* WIP informer2020
* added checking that instantiate works
* added config using gluonTS by kashif
* WIP config
* adding informeConfig. need to remove FeatureEmbedder
* done InformerConfig, but need to change the names
* Done informer model init. working on enc-dec
* added things to address, after reading again enc-dec in the paper
* done modeling - checking initialization work
* added informer to gitignore
* WIP informer2020
* added checking that instantiate works
* added config using gluonTS by kashif
* WIP config
* adding informeConfig. need to remove FeatureEmbedder
* done InformerConfig, but need to change the names
* Done informer model init. working on enc-dec
* added things to address, after reading again enc-dec in the paper
* done modeling - checking initialization work
* moved enc-dec init to InformerEncoder/Decoder init
* added 'init_std' to config, now model init works!
* WIP conversion script, and added code sources
* WIP conversion script: loading original informer pth works
* WIP conversion script: change defaults in the config
* WIP conversion script: supporting Informer input embedding
* WIP conversion script: added parameters for the informer embed
* WIP conversion script: change dim_feedforward=2048
* WIP conversion script: remove unused args for loading checkpoint
* just cleaning up
* DataEmbedding removed, after thinking with Kashif
* working on forward pass
* WIP forward pass: trying to establish working batch for forward pass
* cleaning and finalizing
* adding HF names and docs
* init after cleaning works
* WIP in tests
* added docs for the informer specific args
* fix style
* undo change
* cleaning informer, now need to work only enc-dec
* initial enc-dec classes
* added encoder and decoder
* added todo
* add todos for conv_layers
* added decoder docs from vanilla
* added encoder docs from vanilla
* remove encoder decoder from the original informer
* removed AttentionLayer from the original paper
* removed TriangularCausalMask, same as decoder_attention_mask
* initial sparse attention
* use conv_layers
* fixed test_config test
* fix parenthesis when itearting zip(layers, conv_layers)
* error found in prob attention, added sizes as comments
* fix sizes
* added proposal for q_reduce indexing, and remove unused
* WIP ProbMask, and changed factor=2 for testing
* remove unused libs for this PR for creating the env
* fix checking the attn_weights.size() after bmm
* Q_reduce: changed from torch.gather to simple slicing
* WIP calculate final attn_output
* finish adding v_aggregated, attn_output ready
* changed tgt_len to u in attention_mask, need to fix the size error
* comment attention_mask for encoder, and fix if cond for v_agg
* added ProbMask support (wip), removed old original code
* finished ProbMask 😃
* Revert "remove unused libs for this PR for creating the env"
This reverts commit 11a081e09e.
* fixes
* make style
* fix initial tests
* fix more tests
* dry
* make style
* remove unused files
* style
* added integration tests
* fix num_static_real_features
* fix header
* remove unused function
* fix example
* fix docs
* Update src/transformers/models/informer/configuration_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* Update src/transformers/models/informer/modeling_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* Update src/transformers/models/informer/configuration_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* Update src/transformers/models/informer/configuration_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* Update src/transformers/models/informer/configuration_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* Update src/transformers/models/informer/configuration_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* fixes for reviewer
* use prediction_length from model
* fix style
* fixed informer.mdx
* added to index
* updated readme
* undo
* make fix-copies
* typo
* fix copy
* added Informer to toctree
* in order
* fixed comments
* remove unneeded new lines in docs
* make static real and cat optional
* fix use of distil conv layers
* fixed integration test
* added checkpoint for convlayer
* make fix-copies
* updated from time series model
* make fix-copies
* copy decoder
* fix unit tests
* updated scaling config
* fix integration tests
* IGNORE_NON_TESTED
* IGNORE_NON_AUTO_CONFIGURED
* IGNORE_NON_AUTO_CONFIGURED
* updated check configs
* fix formatting
* undo change from time series
* prediction_length should not be None
* aliign with the blog: prettify ProbSparse and change attention_factor to sampling_factor
* make style
* make fix-copies
* niels CR: update contributed by
* niels CR: update configuration_informer.py
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* niels CR: update kashif -> huggingface
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* niels CR: `sampling_factor` only relevant when `attention_type`=prob
* make style
* fixed U_part: added multiplication by `L_Q`
* fixed bug: remove `is not None` from `if config.distil`
* fixed test: `decoder_seq_length` to `encoder_seq_length` in cross_attentions check
* fix integration tests
* updated model hub
* do not shift as in training
* undo
* fix make-copies
* make fix-copies
* added `if prediction_length is None`
* changed `ProbSparseAttention` to `InformerProbSparseAttention`
* changed `V_sum` -> `v_mean_dim_time`
* changed `ConvLayer` to `InformerConvLayer` and fixed `super()`
* TimeSeriesTansformer->Informer in decoder's Copied from
* more descriptive in ProbSparse
* make style
* fix coped from
* Revert "added `if prediction_length is None`"
This reverts commit b4cbddfa05.
* fixed indent
* use InformerSinusoidalPositionalEmbedding
* make fix-style
* fix from #21860
* fix name
* make fix-copies
* use time series utils
* fix dec num_heads
* docstring
* added time series util doc
* _import_structure
* formatting
* changes from review
* make style
* fix docs
* fix doc
* removed NegativeLogLikelihood
---------
Co-authored-by: Kashif Rasul <kashif.rasul@gmail.com>
Co-authored-by: NielsRogge <48327001+NielsRogge@users.noreply.github.com>
* move generation_*.py src files into generation/*.py
* populate generation.__init__ with lazy loading
* move imports and references from generation.xxx.object to generation.object
