Gemma2: add cache warning (#32279)

* gemma2 fallback to dynamic cache

* Update src/transformers/models/gemma2/modeling_gemma2.py

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

* Update src/transformers/models/gemma2/modeling_gemma2.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* raise error and dont fallback to dynamic cache

* prev will break most forward calls/tests

* Update src/transformers/models/gemma2/modeling_gemma2.py

Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

* update

* fix copies

---------

Co-authored-by: Joao Gante <joaofranciscocardosogante@gmail.com>
Co-authored-by: Arthur <48595927+ArthurZucker@users.noreply.github.com>

docs/source/en/internal/generation_utils.md

@@ -398,6 +398,7 @@ A [`Constraint`] can be used to force the generation to include specific tokens
 
 [[autodoc]] HybridCache
     - update
+    - get_seq_length
     - reset
 
 [[autodoc]] SlidingWindowCache

docs/source/en/model_doc/gemma2.md

@@ -30,6 +30,12 @@ Tips:
 
 - The original checkpoints can be converted using the conversion script `src/transformers/models/Gemma2/convert_Gemma2_weights_to_hf.py` 
 
+<Tip warning={true}>
+
+- Gemma2 uses sliding window attention every second layer, which makes it unsuitable for typical kv caching with [`~DynamicCache`] or tuples of tensors. To enable caching in Gemma2 forward call, you must initialize a [`~HybridCache`] instance and pass it as `past_key_values` to the forward call. Note, that you also have to prepare `cache_position` if the `past_key_values` already contains previous keys and values.
+
+</Tip>
+
 This model was contributed by [Arthur Zucker](https://huggingface.co/ArthurZ), [Pedro Cuenca](https://huggingface.co/pcuenq) and [Tom Arsen]().