GitLab

* [docstring] Fix docstring for `LlamaTokenizer` and `LlamaTokenizerFast`

* [docstring] Fix docstring typo at `LlamaTokenizer` and `LlamaTokenizerFast`

* fix

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

Update docker files to use torch 2.1
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

checkpoint path

* remove from utils

* updated doc string

* only in the model

* Update src/transformers/models/swin/modeling_swin.py
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

* Update src/transformers/models/swin/modeling_swin.py
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

---------
Co-authored-by: Yih-Dar <2521628+ydshieh@users.noreply.github.com>

* [Assistant Generation] Improve enc dec

* save more

* Fix logit processor checks

* Clean

* make style

* fix deprecation

* fix generation test

* Apply suggestions from code review

* fix biogpt

* make style

* copied statement for test files

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

* feat: update callback doc to explain disabling callbacks using report_to

* docs: update report_to docstring

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

* set encoder's PE as non-trainable

* freeze flax

* init sinusoids

* add test for non-trainable embed positions

* simplify TF encoder embed_pos

* revert tf

* clean up

* add sinusoidal init for jax

* make consistent sinusoidal function

* fix dtype

* add default dtype

* use numpy for sinusoids. fix jax

* add sinusoid init for TF

* fix

* use custom embedding

* use specialized init for each impl

* fix sinusoids init. add test for pytorch

* fix TF dtype

* simplify sinusoid init for flax and tf

* add tests for TF

* change default dtype to float32

* add sinusoid test for flax

* Update src/transformers/models/whisper/modeling_flax_whisper.py
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* Update src/transformers/models/whisper/modeling_tf_whisper.py
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

* move sinusoidal init to _init_weights

---------
Co-authored-by: sanchit-gandhi <sanchit@huggingface.co>
Co-authored-by: Sanchit Gandhi <93869735+sanchit-gandhi@users.noreply.github.com>

`jnp.array` is a function, not a type:
https://jax.readthedocs.io/en/latest/_autosummary/jax.numpy.array.html


so it never makes sense to use `jnp.array` in a type annotation. Presumably the intent was to write `jnp.ndarray` aka `jax.Array`.
Co-authored-by: Peter Hawkins <phawkins@google.com>

* fix a typo in flax t5 attention

* fix the typo in flax longt5 attention

* Your commit message here

* fix LlamaConfig docstring

* run make fixup

* fix formatting after review

reformat of the file to prevent script issues

* rerun make fixup after reformat

* control first downsample stride

* reduce first only works for ResNetBottleNeckLayer

* fix param name

* fix style

fix docstrings for vanilla clip

* Fix stale bot

* Comments

* removed donutimageprocessor from objects_to_ignore

* added docstring for donutimageprocessor

* readding donut file

* moved docstring to correct location

fix docstring for CLIPImageProcessor

* fix docstrings for CLIP configs

* black formatted

* fix typos in idefics.md

Two typos found in reviewing this documentation.

1) max_new_tokens=4, is not sufficient to generate "Vegetables" as indicated - you will get only "Veget". (incidentally - some mention of how to select this value might be useful as it seems to change in each example)

2) inputs = processor(prompts, return_tensors="pt").to(device) as inputs need to be on the same device (as they are in all other examples on the page)

* Update idefics.md

Change device to cuda explicitly to match other examples

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

* fix OpenAI GPT, GPT-2 links

* fix Llama2 link

Update test_integration.py

Fixed malapropism clone>copy

* Convert checkpoints

* Update doc test

* Address comment

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

* make sure eos and bos are properly handled for fast tokenizer

* fix code llama as well

* nits

* fix the conversion script as well

* fix failing test

* remove SharedDDP as it was drepracated

* apply review suggestion

* make style

* Oops,forgot to remove the compute_loss context manager in Seq2SeqTrainer.

* remove the unnecessary conditional statement

* keep the logic of IPEX

* clean code

* mix precision setup & make fixup

---------
Co-authored-by: statelesshz <jihuazhong1@huawei.com>

* fix

* fix

* Fix

* Fix

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

* remove unnecessary unsqueeze-squeeze in llama

* correct other models

* fix

* revert gpt_neox_japanese

* fix copie

* fix test

* Update tokenization_code_llama_fast.py

* Update test_tokenization_code_llama.py

* Update test_tokenization_code_llama.py

* 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

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