Support dynamic LoRA loading / unloading in engine/server API (#7446)

python/sglang/srt/managers/tp_worker.py

@@ -30,6 +30,8 @@ from sglang.srt.layers.logits_processor import LogitsProcessorOutput
 from sglang.srt.managers.io_struct import (
     GetWeightsByNameReqInput,
     InitWeightsUpdateGroupReqInput,
+    LoadLoRAAdapterReqInput,
+    UnloadLoRAAdapterReqInput,
     UpdateWeightFromDiskReqInput,
     UpdateWeightsFromDistributedReqInput,
     UpdateWeightsFromTensorReqInput,
@@ -275,3 +277,13 @@ class TpModelWorker:
             recv_req.name, recv_req.truncate_size
         )
         return parameter
+
+    def load_lora_adapter(self, recv_req: LoadLoRAAdapterReqInput):
+        result = self.model_runner.load_lora_adapter(
+            recv_req.lora_name, recv_req.lora_path
+        )
+        return result
+
+    def unload_lora_adapter(self, recv_req: UnloadLoRAAdapterReqInput):
+        result = self.model_runner.unload_lora_adapter(recv_req.lora_name)
+        return result