[diffusion] chore: fix default offload setting for image generation model (#15928)

python/sglang/multimodal_gen/configs/sample/sampling_params.py

@@ -285,7 +285,7 @@ class SamplingParams:
         if pipeline_config.task_type.is_image_gen():
             # settle num_frames
             if not server_args.pipeline_config.allow_set_num_frames():
-                logger.debug(f"num_frames set to 1 for image generation model")
+                logger.debug(f"Setting `num_frames` to 1 for image generation model")
                 self.num_frames = 1
 
         elif self.adjust_frames:

python/sglang/multimodal_gen/runtime/server_args.py

@@ -34,8 +34,6 @@ from sglang.multimodal_gen.utils import FlexibleArgumentParser, StoreBoolean
 
 logger = init_logger(__name__)
 
-ZMQ_TCP_PORT_DELTA = 233
-
 
 def _is_torch_tensor(obj: Any) -> tuple[bool, Any]:
     """Return (is_tensor, torch_module_or_None) without importing torch at module import time."""
@@ -193,11 +191,11 @@ class ServerArgs:
     lora_target_modules: list[str] | None = None
 
     # CPU offload parameters
-    dit_cpu_offload: bool = True
+    dit_cpu_offload: bool | None = None
-    dit_layerwise_offload: bool = False
+    dit_layerwise_offload: bool | None = None
-    text_encoder_cpu_offload: bool = True
+    text_encoder_cpu_offload: bool | None = None
-    image_encoder_cpu_offload: bool = True
+    image_encoder_cpu_offload: bool | None = None
-    vae_cpu_offload: bool = True
+    vae_cpu_offload: bool | None = None
     use_fsdp_inference: bool = False
     pin_cpu_memory: bool = True
 
@@ -266,20 +264,33 @@ class ServerArgs:
 
     def adjust_offload(self):
         if self.pipeline_config.task_type.is_image_gen():
-            logger.info("Turn off all offload for image generation model")
+            logger.info(
-            self.dit_cpu_offload = False
+                "Disabling some offloading (except dit, text_encoder) for image generation model"
+            )
+            if self.dit_cpu_offload is None:
+                self.dit_cpu_offload = True
+            if self.text_encoder_cpu_offload is None:
+                self.text_encoder_cpu_offload = True
+            if self.image_encoder_cpu_offload is None:
+                self.image_encoder_cpu_offload = False
+            if self.vae_cpu_offload is None:
+                self.vae_cpu_offload = False
+        else:
+            self.dit_cpu_offload = True
             self.text_encoder_cpu_offload = True
             self.image_encoder_cpu_offload = True
             self.vae_cpu_offload = True
 
     def __post_init__(self):
-        # Add randomization to avoid race condition when multiple servers start simultaneously
+        # configure logger before use
+        configure_logger(server_args=self)
 
         self.adjust_offload()
 
         if self.attention_backend in ["fa3", "fa4"]:
             self.attention_backend = "fa"
 
+        # Add randomization to avoid race condition when multiple servers start simultaneously
         initial_scheduler_port = self.scheduler_port + random.randint(0, 100)
         self.scheduler_port = self.settle_port(initial_scheduler_port)
         # TODO: remove hard code
@@ -297,8 +308,6 @@ class ServerArgs:
                 raise
         self.check_server_args()
 
-        configure_logger(server_args=self)
-
         # log clean server_args
         try:
             safe_args = _sanitize_for_logging(self, key_hint="server_args")
@@ -733,13 +742,13 @@ class ServerArgs:
 
         if self.ulysses_degree is None:
             self.ulysses_degree = 1
-            logger.info(
+            logger.debug(
                 f"Ulysses degree not set, using default value {self.ulysses_degree}"
             )
 
         if self.ring_degree is None:
             self.ring_degree = 1
-            logger.info(f"Ring degree not set, using default value {self.ring_degree}")
+            logger.debug(f"Ring degree not set, using default value {self.ring_degree}")
 
         if self.ring_degree > 1:
             if self.attention_backend is not None and self.attention_backend != "fa":
@@ -768,7 +777,7 @@ class ServerArgs:
         assert self.dp_size >= 1, "--dp-size must be natural number"
         # NOTE: disable temporarily
         # self.dp_degree = self.num_gpus // self.dp_size
-        logger.info(f"Setting dp_degree to: {self.dp_degree}")
+        logger.debug(f"Setting dp_degree to: {self.dp_degree}")
         if self.dp_size > 1:
             raise ValueError("DP is not yet supported")
 
@@ -957,7 +966,7 @@ def set_global_server_args(server_args: ServerArgs):
     _global_server_args = server_args
 
 
-def get_current_server_args() -> ServerArgs:
+def get_current_server_args() -> ServerArgs | None:
     if _current_server_args is None:
         # in ci, usually when we test custom ops/modules directly,
         # we don't set the sgl_diffusion config. In that case, we set a default
@@ -967,7 +976,7 @@ def get_current_server_args() -> ServerArgs:
     return _current_server_args
 
 
-def get_global_server_args() -> ServerArgs:
+def get_global_server_args() -> ServerArgs | None:
     if _global_server_args is None:
         # in ci, usually when we test custom ops/modules directly,
         # we don't set the sgl_diffusion config. In that case, we set a default

python/sglang/multimodal_gen/test/slack_utils.py

@@ -50,11 +50,10 @@ except Exception as e:
 
 def _get_status_message(run_id, current_case_id, thread_messages=None):
     date_str = datetime.now().strftime("%d/%m")
-    base_header = f"""🧵 for nightly test of {date_str}*
+    base_header = f"""🧵 for nightly test of {date_str}
 *Git Revision:* {get_git_commit_hash()}
 *GitHub Run ID:* {run_id}
 *Total Tasks:* {len(ALL_CASES)}
-
 """
 
     if not ALL_CASES: