Merge pull request #5515 from Edenzzzz/fix_layout_convert

Fix layout convertor caching

colossalai/tensor/d_tensor/layout_converter.py

@@ -440,7 +440,10 @@ class LayoutConverter(metaclass=SingletonMeta):
         total_steps = 0
         transform_path = []
         comm_action_sequence: List[CommSpec] = []
-        spec_pairs = (str(source_spec.sharding_sequence), str(target_spec.sharding_sequence))
+
+        src_shape = source_layout.get_sharded_shape_per_device()
+        dst_shape = target_layout.get_sharded_shape_per_device()
+        spec_pairs = ((str(source_spec.sharding_sequence), src_shape), (str(target_spec.sharding_sequence), dst_shape))
 
         if spec_pairs in self.cached_solution:
             # Solution Cache hit

tests/test_tensor/test_dtensor/test_layout_converter.py

@@ -123,8 +123,15 @@ def check_layout_converting(rank, world_size, port):
     assert comm_action_sequence[2].logical_process_axis == 1
 
     # checkout chached_spec_pairs_transform_path
-    assert layout_converter.cached_solution[("[R, S01, R]", "[S01, R, R]")][0] == transform_path
+    src_shape = source_layout.get_sharded_shape_per_device()
-    assert layout_converter.cached_solution[("[R, S01, R]", "[S01, R, R]")][1] == comm_action_sequence
+    dst_shape = target_layout.get_sharded_shape_per_device()
+    assert (
+        layout_converter.cached_solution[(("[R, S01, R]", src_shape), ("[S01, R, R]", dst_shape))][0] == transform_path
+    )
+    assert (
+        layout_converter.cached_solution[(("[R, S01, R]", src_shape), ("[S01, R, R]", dst_shape))][1]
+        == comm_action_sequence
+    )
 
     comm_cost = layout_converter.get_total_comm_cost(source_layout, target_layout)