Multiple llama4 fixe (#37353)

* update for fixes

* more fixes

* fuxix dynamic cache?

* style

* fix both traiining and generating. Eager seems alright

* dynamic does not work

* fix most cases, use_cache or not, eager or not, no default cache (ex: not training but you want to get cache states)

* should be final fixes

* fix more stuff no cat

* style

* fix

* style

* final sytle

* qualityeioiwhjfaopsejdpofqsdjkfjha;wesdhgfkjlqsw.denghjkaswednkgs

* fix

* revert

utils/check_config_attributes.py

@@ -244,6 +244,7 @@ SPECIAL_CASES_TO_ALLOW = {
         "output_router_logits",
         "router_aux_loss_coef",
         "router_jitter_noise",
+        "cache_implementation",
     ],
     "Llama4VisionConfig": ["multi_modal_projector_bias", "norm_eps"],
 }

utils/check_docstrings.py

@@ -580,6 +580,7 @@ OBJECTS_TO_IGNORE = [
     "ZeroShotClassificationPipeline",
     "ZeroShotImageClassificationPipeline",
     "ZeroShotObjectDetectionPipeline",
+    "Llama4TextConfig",
 ]
 
 # Supported math operations when interpreting the value of defaults.