Feature/reduce-decoder-mem-usage

CHANGELOG.md

@@ -36,6 +36,7 @@ Keep it human-readable, your future self will thank you!
 - Update copyright notice
 - Fix `__version__` import in init
 - Fix missing copyrights [#71](https://github.com/ecmwf/anemoi-models/pull/71)
+- Reduced memory usage when using chunking in the mapper [#84](https://github.com/ecmwf/anemoi-models/pull/84)
 
 ### Removed
 

src/anemoi/models/layers/block.py

@@ -512,18 +512,16 @@ def forward(
             edge_attr_list, edge_index_list = sort_edges_1hop_chunks(
                 num_nodes=size, edge_attr=edges, edge_index=edge_index, num_chunks=num_chunks
             )
+            out = torch.zeros((x[1].shape[0], self.num_heads, self.out_channels_conv), device=x[1].device)
             for i in range(num_chunks):
-                out1 = self.conv(
+                out += self.conv(
                     query=query,
                     key=key,
                     value=value,
                     edge_attr=edge_attr_list[i],
                     edge_index=edge_index_list[i],
                     size=size,
                 )
-                if i == 0:
-                    out = torch.zeros_like(out1, device=out1.device)
-                out = out + out1
         else:
             out = self.conv(query=query, key=key, value=value, edge_attr=edges, edge_index=edge_index, size=size)