diff --git a/src/diffusers/hooks/context_parallel.py b/src/diffusers/hooks/context_parallel.py
index 1310b20c5c11..e8bd4c5bb28f 100644
--- a/src/diffusers/hooks/context_parallel.py
+++ b/src/diffusers/hooks/context_parallel.py
@@ -57,6 +57,9 @@ def _get_parameter_from_args_kwargs(self, identifier: str, args=(), kwargs=None)
             index = self.cached_parameter_indices.get(identifier, None)
             if index is None:
                 raise ValueError(f"Parameter '{identifier}' not found in cached indices.")
+            if index >= len(args):
+                # Not passed positionally or as a kwarg; the forward will use its default value.
+                return None, False, index
             return args[index], False, index
 
         if self._cls is None:
@@ -72,7 +75,8 @@ def _get_parameter_from_args_kwargs(self, identifier: str, args=(), kwargs=None)
         index = self.cached_parameter_indices[identifier]
 
         if index >= len(args):
-            raise ValueError(f"Expected {index} arguments but got {len(args)}.")
+            # Not passed positionally or as a kwarg; the forward will use its default value.
+            return None, False, index
 
         return args[index], False, index
 
diff --git a/src/diffusers/models/attention_dispatch.py b/src/diffusers/models/attention_dispatch.py
index 9414c151fd67..a07ca93ebc06 100644
--- a/src/diffusers/models/attention_dispatch.py
+++ b/src/diffusers/models/attention_dispatch.py
@@ -2395,6 +2395,10 @@ def forward(
             mask_list = [torch.empty_like(attn_mask) for _ in range(world_size)]
             dist.all_gather(mask_list, attn_mask, group=group)
             attn_mask = torch.cat(mask_list, dim=-1)
+        if attn_mask is not None and attn_mask.ndim == 4 and attn_mask.shape[1] == H:
+            # QKV carry H // world_size heads after the all-to-all; shard the mask's head axis to match.
+            local_rank = _parallel_config.context_parallel_config._ulysses_local_rank
+            attn_mask = attn_mask.chunk(world_size, dim=1)[local_rank]
 
         out = forward_op(
             ctx,
diff --git a/src/diffusers/models/transformers/transformer_ltx2.py b/src/diffusers/models/transformers/transformer_ltx2.py
index 465408d94693..26c36c3a5295 100644
--- a/src/diffusers/models/transformers/transformer_ltx2.py
+++ b/src/diffusers/models/transformers/transformer_ltx2.py
@@ -1091,18 +1091,6 @@ class LTX2VideoTransformer3DModel(
     _supports_gradient_checkpointing = True
     _skip_layerwise_casting_patterns = ["norm"]
     _repeated_blocks = ["LTX2VideoTransformerBlock"]
-    _cp_plan = {
-        "": {
-            "hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
-            "encoder_hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
-            "encoder_attention_mask": ContextParallelInput(split_dim=1, expected_dims=2, split_output=False),
-        },
-        "rope": {
-            0: ContextParallelInput(split_dim=1, expected_dims=3, split_output=True),
-            1: ContextParallelInput(split_dim=1, expected_dims=3, split_output=True),
-        },
-        "proj_out": ContextParallelOutput(gather_dim=1, expected_dims=3),
-    }
 
     @register_to_config
     def __init__(
@@ -1318,6 +1306,37 @@ def __init__(
 
         self.gradient_checkpointing = False
 
+        # RoPE modules return (cos, sin) shaped (B, T, D) for rope_type="interleaved" and (B, H, T, D // 2)
+        # for rope_type="split", so the sequence dim to shard depends on the config.
+        if rope_type == "split":
+            rope_plan = {
+                0: ContextParallelInput(split_dim=2, expected_dims=4, split_output=True),
+                1: ContextParallelInput(split_dim=2, expected_dims=4, split_output=True),
+            }
+        else:
+            rope_plan = {
+                0: ContextParallelInput(split_dim=1, expected_dims=3, split_output=True),
+                1: ContextParallelInput(split_dim=1, expected_dims=3, split_output=True),
+            }
+        self._cp_plan = {
+            "": {
+                "hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
+                "timestep": ContextParallelInput(split_dim=1, expected_dims=2, split_output=False),
+                "audio_hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
+                "audio_timestep": ContextParallelInput(split_dim=1, expected_dims=2, split_output=False),
+                "encoder_hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
+                "encoder_attention_mask": ContextParallelInput(split_dim=1, expected_dims=2, split_output=False),
+                "audio_encoder_hidden_states": ContextParallelInput(split_dim=1, expected_dims=3, split_output=False),
+                "audio_encoder_attention_mask": ContextParallelInput(split_dim=1, expected_dims=2, split_output=False),
+            },
+            "rope": rope_plan,
+            "audio_rope": rope_plan,
+            "cross_attn_rope": rope_plan,
+            "cross_attn_audio_rope": rope_plan,
+            "proj_out": ContextParallelOutput(gather_dim=1, expected_dims=3),
+            "audio_proj_out": ContextParallelOutput(gather_dim=1, expected_dims=3),
+        }
+
     @apply_lora_scale("attention_kwargs")
     def forward(
         self,