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

python/sglang/srt/entrypoints/http_server.py

@@ -72,6 +72,7 @@ from sglang.srt.managers.io_struct import (
     GenerateReqInput,
     GetWeightsByNameReqInput,
     InitWeightsUpdateGroupReqInput,
+    LoadLoRAAdapterReqInput,
     OpenSessionReqInput,
     ParseFunctionCallReq,
     ProfileReqInput,
@@ -80,6 +81,7 @@ from sglang.srt.managers.io_struct import (
     SeparateReasoningReqInput,
     SetInternalStateReq,
     SlowDownReqInput,
+    UnloadLoRAAdapterReqInput,
     UpdateWeightFromDiskReqInput,
     UpdateWeightsFromDistributedReqInput,
     UpdateWeightsFromTensorReqInput,
@@ -595,6 +597,40 @@ async def slow_down(obj: SlowDownReqInput, request: Request):
         return _create_error_response(e)
 
 
+@app.api_route("/load_lora_adapter", methods=["POST"])
+async def load_lora_adapter(obj: LoadLoRAAdapterReqInput, request: Request):
+    """Load a new LoRA adapter without re-launching the server."""
+    result = await _global_state.tokenizer_manager.load_lora_adapter(obj, request)
+
+    if result.success:
+        return ORJSONResponse(
+            result,
+            status_code=HTTPStatus.OK,
+        )
+    else:
+        return ORJSONResponse(
+            result,
+            status_code=HTTPStatus.BAD_REQUEST,
+        )
+
+
+@app.api_route("/unload_lora_adapter", methods=["POST"])
+async def unload_lora_adapter(obj: UnloadLoRAAdapterReqInput, request: Request):
+    """Load a new LoRA adapter without re-launching the server."""
+    result = await _global_state.tokenizer_manager.unload_lora_adapter(obj, request)
+
+    if result.success:
+        return ORJSONResponse(
+            result,
+            status_code=HTTPStatus.OK,
+        )
+    else:
+        return ORJSONResponse(
+            result,
+            status_code=HTTPStatus.BAD_REQUEST,
+        )
+
+
 @app.api_route("/open_session", methods=["GET", "POST"])
 async def open_session(obj: OpenSessionReqInput, request: Request):
     """Open a session, and return its unique session id."""