[diffusion] fix: avoid saving output for warmup requests (#18867)

python/sglang/multimodal_gen/runtime/managers/scheduler.py

@@ -198,7 +198,6 @@ class Scheduler:
                         prompt="",
                         negative_prompt="",
                         image_path=[input_path],
-                        is_warmup=True,
                     )
                 else:
                     req = Req(
@@ -206,8 +205,8 @@ class Scheduler:
                         width=width,
                         height=height,
                         prompt="",
-                        is_warmup=True,
                     )
+                req.set_as_warmup()
                 self.waiting_queue.append((None, req))
             # if server is warmed-up, set this flag to avoid req-based warmup
             self.warmed_up = True

python/sglang/multimodal_gen/runtime/pipelines_core/schedule_batch.py

@@ -252,6 +252,8 @@ class Req:
 
     def set_as_warmup(self):
         self.is_warmup = True
+        self.save_output = False
+        self.suppress_logs = True
         self.extra["cache_dit_num_inference_steps"] = self.num_inference_steps
         self.num_inference_steps = 1