Sharding across width

Author: prishajain1

src/maxdiffusion/models/wan/autoencoder_kl_wan.py

@@ -137,7 +137,9 @@ def __call__(self, x: jax.Array, cache_x: Optional[jax.Array] = None, idx=-1) ->
     if self.mesh is not None:
       # (B, D, H, W, C)
       if x_padded.shape[0] % self.mesh.shape['data'] == 0:
-        x_padded = with_sharding_constraint(x_padded, PartitionSpec('data', None, None, None, None))
+        x_padded = with_sharding_constraint(x_padded, PartitionSpec('data', None, None, 'fsdp', None))
+      else:
+        x_padded = with_sharding_constraint(x_padded, PartitionSpec(None, None, None, 'fsdp', None))
 
     out = self.conv(x_padded)
     return out