🧹 Remove deprecated RotaryEmbedding parts in the Attention layers (#34858)

* update

* style

* fix missing args

* remove last trace of old rope classes

* remove deprecated copied from

* fix copies

* trigger CIs

* post rebase clean-up

* reverse mistral

* cleanup after dropping commits

* Add comment

tests/models/falcon/test_modeling_falcon.py

@@ -50,8 +50,6 @@ if is_torch_available():
         FalconModel,
     )
     from transformers.models.falcon.modeling_falcon import (
-        FalconDynamicNTKScalingRotaryEmbedding,
-        FalconLinearScalingRotaryEmbedding,
         FalconRotaryEmbedding,
     )
 
@@ -484,11 +482,12 @@ class FalconModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix
 
         # Sanity check linear RoPE scaling
         # New position "x" should match original position with index "x/scaling_factor"
-        linear_scaling_rope = FalconLinearScalingRotaryEmbedding(
+        linear_scaling_rope = FalconRotaryEmbedding(
             head_dim,
             max_position_embeddings=config.max_position_embeddings,
             base=config.rope_theta,
             scaling_factor=scaling_factor,
+            rope_type="linear",
         ).to(torch_device)
         linear_cos_short, linear_sin_short = linear_scaling_rope(x, position_ids_short)
         linear_cos_long, linear_sin_long = linear_scaling_rope(x, position_ids_long)
@@ -502,11 +501,12 @@ class FalconModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMix
         # Sanity check Dynamic NTK RoPE scaling
         # Scaling should only be observed after a long input is fed. We can observe that the frequencies increase
         # with scaling_factor (or that `inv_freq` decreases)
-        ntk_scaling_rope = FalconDynamicNTKScalingRotaryEmbedding(
+        ntk_scaling_rope = FalconRotaryEmbedding(
             head_dim,
             max_position_embeddings=config.max_position_embeddings,
             base=config.rope_theta,
             scaling_factor=scaling_factor,
+            rope_type="dynamic",
         ).to(torch_device)
         ntk_cos_short, ntk_sin_short = ntk_scaling_rope(x, position_ids_short)
         ntk_cos_long, ntk_sin_long = ntk_scaling_rope(x, position_ids_long)