Bump protobuf in /examples/onnxruntime/training/language-modeling

Bumps [protobuf](https://github.com/protocolbuffers/protobuf) from 3.20.2 to 4.25.8.
- [Release notes](https://github.com/protocolbuffers/protobuf/releases)
- [Changelog](https://github.com/protocolbuffers/protobuf/blob/main/protobuf_release.bzl)
- [Commits](protocolbuffers/protobuf@v3.20.2...v4.25.8)

---
updated-dependencies:
- dependency-name: protobuf
  dependency-version: 4.25.8
  dependency-type: direct:production
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* Move task register in onnx exporters

* fix

* remove unused

* fix

* fix

* use model_types as god intended for them

* use the names transformers has given them

* fix perceiver

* fix perceiver text vision distinction

* fix other models

* add reference to official model types in transformers

* use automated script

* data2vec

* remove duplicate

* fix

* fix

* runscript with latest transformers

* add fix chinese clip

* apply script

* fix

---------

Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>
Co-authored-by: Ella Charlaix <ella@huggingface.co>

* add type checking

* tflite config exporters

* add exporters config

* fix style

* fix

* fix

* remove constant

* fix style

* move

* rm comments

* fix masked-im output names

* style

* rm comment

* fix doc generation

* task attribute already set

* model patcher

* add model patcher

* add processor

* move patching specs to onnx config

* rename

* add comment

* fix

* style

* rename

* style

* remove todo comments

---------

Co-authored-by: IlyasMoutawwakil <moutawwakil.ilyas.tsi@gmail.com>

* remove unused method

* remove unused

…argument in `quantize_model`' while loading Qwen VL GPTQ model (#2295)

Updated BLOCK_PATTERNS

remove timm from required dependencies

Co-authored-by: Abdennacer Badaoui <abdennacerbadaoui@MacBook-Air-de-Abdennacer.local>

Revert "Adding DummyInputGenerator for VisualBert (#2303)"

This reverts commit a4152c0.

* add dummy input generator

* fix style

* add default shapes

* Apply suggestions from code review

---------

Co-authored-by: Ilyas Moutawwakil <57442720+IlyasMoutawwakil@users.noreply.github.com>

fix tflite

* deprecate tflite

* bt and ort training

…2319)

* test

* fix seq2seq patched sdpa

* patch qwen3_moe, out_attentions, and eager_mask

* fix

* use optimum model

* editable subpackages

* Apply suggestions from code review

* smollm3 support

* deprecate tensorflow onnx export and add smollm3 to export tests

* write a more general sdpa_mask without vmap that's also vectorized and broadcastable

* better and more generic sdpa_mask_without_vmap implementation

* style and fix

* fix

* patch find_packed_sequence_indices as it's untraceable

* fix

* fix

* revert tests removal until refactor

* fix temporary hub repo import

* fix

* fix external data tests on windows

* update phi and phi3 min version

* condition modernbert optimization test

* get back old (pre 4.44) bloom modeling support and remove the need for normalized config in ort modeling

* fix test was using hardcoded architecture

* unparallelize test that uses remote code

* support older versions of mpt and phi (4.36)

* remove parallelism from slow tests

* fix vision to text pipelines test

* more specific version handling for find_packed_sequence_indices

* fix

* Propagate library_name parameter in from_pretrained to export

Required to avoid automatic inferring of the library_name

* Use class attribute for ORTModel instead

Under modeling_diffusion it looks like ORTModel isn't used

* Add test case

* Update optimum/onnxruntime/modeling_ort.py

---------

Co-authored-by: Ella Charlaix <80481427+echarlaix@users.noreply.github.com>

…, gpt_bigcode multi-query,.. (#2326)

* test left-padded batched inference

* demonstrate batched tex generation failure

* fix remote code

* fix

* fix position_ids generation inside ORTModelForCausalLM class

* it works until transformers 4.52 -_-

* now run with latest transformers

* bolean 4D mask is actually not supported by torch onnx exporter

* only test generation with batched inputs, for logits are a bit off because of transformers using boolean mask

* boolean mask safe softmax batched inference

* style

* use old typing

* don't do unnecessary patching

* try to avoid spamming the hub for an image

* update min transformers version

* better and direct torch patching

* more batched generation special cases

* style

* initialize the il image instead of downloading it

* use random pil image

* test different versions of transformers in fast tests

* fix

* revert diffusers changes for now

* mask padding kv cache as well

* fix masking for old bloom

* use constant image to image loading errors

* style

* test diffusers in series to avoid runner dying

* fix

* cleanup and some comments

* fix and test falcon alibi

* style

* fix, support and test multi_query=False as well

* only apply masked testing for transformers version previous to 4.39

* Update optimum/onnxruntime/modeling_decoder.py

* use text decoder position ids onnx config but test its sync with list

* fix opt

* style

* fix sdpa without overriting torch onnx exporter

* use inplace op ;-;

* fix st test

* patch directly in onnx because patch needs to happen after softmax