fix dual stream cross attention masking bug

Author: prishajain1

src/maxdiffusion/models/ltx2/transformer_ltx2.py

@@ -1189,8 +1189,8 @@ def scan_fn(carry, block_mask_and_id):
             audio_rotary_emb=audio_rotary_emb,
             ca_video_rotary_emb=video_cross_attn_rotary_emb,
             ca_audio_rotary_emb=audio_cross_attn_rotary_emb,
-            a2v_cross_attention_mask=encoder_attention_mask,
-            v2a_cross_attention_mask=audio_encoder_attention_mask,
+            a2v_cross_attention_mask=None,
+            v2a_cross_attention_mask=None,
             perturbation_mask=mask,
             modality_mask=modality_mask,
         )
@@ -1235,8 +1235,8 @@ def scan_fn(carry, block_mask_and_id):
               ca_audio_rotary_emb=audio_cross_attn_rotary_emb,
               encoder_attention_mask=encoder_attention_mask,
               audio_encoder_attention_mask=audio_encoder_attention_mask,
-              a2v_cross_attention_mask=encoder_attention_mask,
-              v2a_cross_attention_mask=audio_encoder_attention_mask,
+              a2v_cross_attention_mask=None,
+              v2a_cross_attention_mask=None,
               perturbation_mask=mask,
           )