Bart: new cache format (#35314)

* bart compile

* add mbart

* some more models touched by fix-copies

* more

* more models

* even more models

* fix copies

* fix tests

* fix copies

* fix

* biogpt accepts position ids now (breaking?)

* fix failing non-slow tests

* fix some tests

* should not be removed

* small update

* Update src/transformers/models/bart/modeling_bart.py

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

* update for last `main`

* fix copies

* clone `update_causal_mask` from llama

* tmp

* fixup

* why? how?

* fix bart tests

* dont skip test

* address comments

* fix tests

* fix

* fixup and delete the file

---------

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>

tests/generation/test_utils.py

@@ -2144,6 +2144,7 @@ class GenerationTesterMixin:
             compile_config._compile_all_devices = True  # force compilation (e.g. fast CI, CPU)
 
             generation_kwargs = {
+                "use_cache": True,
                 "do_sample": False,
                 "max_new_tokens": 5,
                 "return_dict_in_generate": True,