GitLab

* initial implementation of flash attention for gptj

* modify flash attention and overwrite test_flash_attn_2_generate_padding_right

* update flash attention support list

* remove the copy line in the `CodeGenBlock`

* address copy mechanism

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

* Add GPTJ attention classes

* add expected outputs in the gptj test

* Ensure repo consistency with 'make fix-copies'

---------
Co-authored-by: Younes Belkada <49240599+younesbelkada@users.noreply.github.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

DeepSpeed: hardcode `torch.arange` dtype on `float` usage to avoid incorrect initialization (#28760)

* normalize reverse indexing for causal lm sequence classifiers

* normalize reverse indexing for causal lm sequence classifiers

* normalize reverse indexing for causal lm sequence classifiers

* use modulo instead

* unify modulo-based sequence lengths

support onnx for causal lm sequence classification

* fix

* more fixes

* fix other models

* fix long t5

* use `gradient_checkpointing_func` instead

* fix copies

* set `gradient_checkpointing_func` as a private attribute and retrieve previous behaviour

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

* replace it with `is_gradient_checkpointing_set`

* remove default

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

* fixup

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

* v1

* fix

* remove `create_custom_forward`

* fixup

* fixup

* add test and fix all failing GC tests

* remove all remaining `create_custom_forward` methods

* fix idefics bug

* fixup

* replace with `__call__`

* add comment

* quality

* fix key dtype in gptj and codegen

* delay the key cast to a later point

* fix

In assisted decoding, pass model_kwargs to model's forward call (fix prepare_input_for_generation in all models) (#25242)

* In assisted decoding, pass model_kwargs to model's forward call

Previously, assisted decoding would ignore any additional kwargs
that it doesn't explicitly handle. This was inconsistent with other
generation methods, which pass the model_kwargs through
prepare_inputs_for_generation and forward the returned dict to the
model's forward call.

The prepare_inputs_for_generation method needs to be amended in all
models, as previously it only kept the last input ID when a past_key_values
was passed.

* Improve variable names in _extend_attention_mask

* Refactor extending token_type_ids into a function

* Replace deepcopy with copy to optimize performance

* Update new persimmon model with llama changes for assisted generation

* Update new mistral model for assisted generation with prepare_inputs_for_generation

* Update position_ids creation in falcon prepare_inputs_for_generation to support assisted generation

* Remove unnecessary `view` of `position_ids` in `modeling_llama`

When `position_ids` is `None`, its value is generated using
`torch.arange`, which creates a tensor of size `(seq_length +
past_key_values_length) - past_key_values_length = seq_length`. The
tensor is then unsqueezed, resulting in a tensor of shape `(1,
seq_length)`. This means that the last `view` to a tensor of shape
`(-1, seq_length)` is a no-op.

This commit removes the unnecessary view.

* Remove no-op `view` of `position_ids` in rest of transformer models

* Add attention mask and pad token warning to many of the models

* Remove changes under examples/research_projects

These files are not maintained by HG.

* Skip the warning check during torch.fx or JIT tracing

* Switch ordering for the warning and input shape assignment

This ordering is a little cleaner for some of the cases.

* Add missing line break in one of the files

* Loosen output shape restrictions on GPT-style models

* Use more self-explanatory variables

* Revert "Use more self-explanatory variables"

This reverts commit 5fd9ab39119558b7e750f61aa4a19014dccc5ed5.

* support left padding

* nit

* Update src/transformers/models/gpt_neox/modeling_gpt_neox.py

* Update src/transformers/models/gpt_neox/modeling_gpt_neox.py

* Preliminary work on some models

* Fix test load missing and make sure nonpersistent buffers are tested

* Always ignore nonpersistent buffers if in state_dict

* Treat models

* More models

* Treat remaining models

* Fix quality

* Fix tests

* Remove draft

* This test is not needed anymore

* Fix copies

* Fix last test

* Newly added models

* Fix last tests

* Address review comments

Revert "Fix gradient checkpointing + fp16 autocast for most models (#24247)"

This reverts commit 285a4801.

* fix gc bug

* continue PoC on OPT

* fixes

* :exploding_head:

* fix tests

* remove pytest.mark

* fixup

* forward contrib credits from discussions

* forward contrib credits from discussions

* reverting changes on untouched files.

---------
Co-authored-by: zhaoqf123 <zhaoqf123@users.noreply.github.com>
Co-authored-by: 7eu7d7 <7eu7d7@users.noreply.github.com>

* First test

* Add info for all models

* style

* Repo consistency

* Fix last model and cleanup prints

* Repo consistency

* Use consistent function for detecting tied weights

Signed-off-by: Wang, Yi A <yi.a.wang@intel.com>

* fix #23136

* better fix

* same fix for `masked_bias`

* first draft - gives index error in question_answering.py

* maturing

* no labels

* pipeline should know about QA

* fixing checks

* formatting

* fixed docstring

* initial commit

* formatting

* adding the class to many places

* towards less unhappy checks

* nearly there

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

* avoid error

* moving to device of star/end_logits

---------
Co-authored-by: Prof. Peter Schneider-Kamp <jps@ordbogen.com>
Co-authored-by: amyeroberts <22614925+amyeroberts@users.noreply.github.com>

* moved labels to the same device as logits for OTP model

* moved labels to the same device as logits for CODEGEN model

* Update modeling_codegen.py

* moved labels to the same device as logits for gptj and pix2struct model

* Update modeling_pix2struct.py

* [gptj] support older pytorch version

* contributor

* contributor

* make copies

---------
Co-authored-by: Michael Wyatt <michaelwyatt@microsoft.com>
Co-authored-by: Nick Hill <nickhill@us.ibm.com>

* Revert "[GPT-J] add deprecation warning (#21869)"

This reverts commit fb76994c.

* Fix position embeddings for GPT-J and CodeGen

* Address review comments from @gante

* Fix "Copied from" comment referencing wrong function

* Fix copy/paste mistake

* Fix training path

* Hopefully make torch.fx happy

* Move position_ids long cast

* Revert "Hopefully make torch.fx happy"

This reverts commit e41a6f4cad3ff441124c7457b19cfb630d4ca025.

* Changes to help with torch.fx tracing

* Linter fix

* Correct position_ids tensor type hint

* Work-around torch.fx tracing issue

* Get the changes to work with torch.fx

* Address review comment from @michaelbenayoun

* Another small adjustment

* Add explanatory comment; small code tidyup

* add deprecation warning

* remove pos ids from args docstirng

* fix failing test

* If applied, this commit fixes generate bug in gptj

* Remove extra same code block

* formatting and test fix

* Conflict fix and declaration error fix

---------
Co-authored-by: Sylvain Gugger <35901082+sgugger@users.noreply.github.com>

* logger.warning_once

* style

* uint8 -> bool

* fix copies

* style

* update test modeling commen when checking attention buffers

* style

* use logical not on random mask instead of subtraction with 1

* remove torch uint8

* quality

* remove modified modeling utils

* Update based on review
Co-authored-by: sgugger <sylvain.gugger@gmail.com>

---------
Co-authored-by: sgugger <sylvain.gugger@gmail.com>

* fix past renamed to past_key_value

* update more `past`that were ski^êd

* fixup

* remove changes made to rag

* refactor `_reorder_cache` to use `past_key_values`

* fix git `prepare_inputs_for_generation` to pass tests when false is needed in use_cache

* Deprecate parallelize API

* Add documentation

* Fix copies

* Result of black 23.1

* Update target to Python 3.7

* Switch flake8 to ruff

* Configure isort

* Configure isort

* Apply isort with line limit

* Put the right black version

* adapt black in check copies

* Fix copies

* Clean all models

* Style

* Last to remove

* address review comments

* Address review comments

Replace the checkpoint - the current one has shape issue
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* Add disclaimer for necessary fake models

* Address review comments

* Use for GPT-NeoX as well

* start cleanup

* more updates

* more models are affected

* more updates

* update generation utils

* style

* revert change that removed reorder cachce

* update generation utils

* style

* style

* remove reorder cache

* fix 1

* fix 2

* fix 3

* fix 4
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

* fix incorrect comments about atten mask

* typo

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