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

* fix gc bug

* continue PoC on OPT

* fixes

* 🤯

* 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>

tests/models/big_bird/test_modeling_big_bird.py

@@ -609,6 +609,12 @@ class BigBirdModelTest(ModelTesterMixin, PipelineTesterMixin, unittest.TestCase)
         config_and_inputs = self.model_tester.prepare_config_and_inputs()
         self.model_tester.create_and_check_for_change_to_full_attn(*config_and_inputs)
 
+    @unittest.skip(
+        reason="The model does not support GC + autocast + fp16: https://github.com/huggingface/transformers/pull/24247"
+    )
+    def test_training_gradient_checkpointing_autocast(self):
+        pass
+
     # overwrite from common in order to skip the check on `attentions`
     def check_pt_flax_outputs(self, fx_outputs, pt_outputs, model_class, tol=1e-5, name="outputs", attributes=None):
         # `bigbird_block_sparse_attention` in `FlaxBigBird` returns `attention_probs = None`, while in PyTorch version,