[LoRA] 1/n Per-rank tensor serialization for load_lora_adapter_from_tensors under dp_size > 1 (#32580)

test/registered/rl/test_lora_load_from_tensor.py

@@ -342,9 +342,11 @@ class TestLoRALoadFromTensor(CustomTestCase):
         }
         serialized = MultiprocessingSerializer.serialize(bucket_dict, output_str=True)
 
+        # flattened_bucket callers pass one serialized copy per TP rank, same
+        # as Engine.update_weights_from_tensor.
         result = self.engine.load_lora_adapter_from_tensors(
             lora_name="self_cognition_Alice_flattened",
-            tensors=serialized,
+            tensors=[serialized],
             config_dict=self.lora_config_dict,
             load_format="flattened_bucket",
         )

test/registered/unit/managers/test_msgpack_ipc_roundtrip.py

@@ -126,7 +126,7 @@ REGISTRY_TYPE_INSTANCES = {
     "LoadLoRAAdapterFromTensorsReqInput": LoadLoRAAdapterFromTensorsReqInput(
         lora_name="adapter",
         config_dict={"r": 8, "lora_alpha": 16, "target_modules": ["q_proj", "v_proj"]},
-        serialized_tensors="",
+        serialized_named_tensors=[b"tp0-bytes", b"tp1-bytes"],
         added_tokens_config={"<extra>": 32000},
     ),
     "DumperControlReqInput": DumperControlReqInput(method="start", body={"k": "v"}),