[diffusion] UX: reduce hot-path server log noise (#37804)

python/sglang/multimodal_gen/runtime/disaggregation/orchestrator.py

@@ -650,7 +650,7 @@ class DiffusionServer:
                 )
             except ValueError:
                 pass
-            logger.info(
+            logger.debug(
                 "GLM distributed mode dispatched request with %d output(s) "
                 "to denoiser[%d]",
                 request.req.num_outputs_per_prompt,

python/sglang/multimodal_gen/runtime/pipelines_core/stages/denoising.py

@@ -706,7 +706,7 @@ class DenoisingStage(PipelineStage, RolloutDenoisingMixin):
 
         self._quality_fusions_mounted = want
         for description in sorted(mounted_fusions):
-            logger.info("Mounted %s for quality=%s", description, quality)
+            logger.debug("Mounted %s for quality=%s", description, quality)
 
     def _cache_dit_dual_model_name(self) -> str:
         return "wan2.2"

python/sglang/multimodal_gen/runtime/pipelines_core/stages/model_specific_stages/minimax_h3/stages/denoising.py

@@ -371,7 +371,7 @@ def _build_cube_attn_metadata(
         num_steps=num_steps,
         device=device,
     )
-    logger.info(
+    logger.debug(
         "cube sparse attention enabled: local_cube_size=%s "
         "topk_ratio_list(len=%d, min=%.4f, max=%.4f)",
         list(local_cube_size),

python/sglang/multimodal_gen/runtime/vla/cuda_graph.py

@@ -109,6 +109,18 @@ class _BoundedCaptureCache:
             or self.evict_on_miss
         )
 
+    def _evict_lru(self) -> None:
+        evicted_signature, evicted = self.entries.popitem(last=False)
+        self._release(evicted)
+        self.evictions += 1
+        logger.debug(
+            "Evicted VLA %s CUDA graph for signature %s (entries=%d/%d)",
+            self.name,
+            evicted_signature,
+            len(self.entries),
+            self.max_entries,
+        )
+
     def prepare_admission(self, signature: Any) -> None:
         if (
             signature in self.entries
@@ -116,16 +128,7 @@ class _BoundedCaptureCache:
             or not self.evict_on_miss
         ):
             return
-        evicted_signature, evicted = self.entries.popitem(last=False)
+        self._evict_lru()
-        self._release(evicted)
-        self.evictions += 1
-        logger.info(
-            "Evicted VLA %s CUDA graph for signature %s (entries=%d/%d)",
-            self.name,
-            evicted_signature,
-            len(self.entries),
-            self.max_entries,
-        )
 
     def put(self, signature: Any, entry: Any) -> bool:
         if self.max_entries == 0 or not self.can_admit(signature):
@@ -138,16 +141,7 @@ class _BoundedCaptureCache:
         self.captures += 1
 
         if len(self.entries) > self.max_entries:
-            evicted_signature, evicted = self.entries.popitem(last=False)
+            self._evict_lru()
-            self._release(evicted)
-            self.evictions += 1
-            logger.info(
-                "Evicted VLA %s CUDA graph for signature %s (entries=%d/%d)",
-                self.name,
-                evicted_signature,
-                len(self.entries),
-                self.max_entries,
-            )
         return True
 
     def discard(self, signature: Any) -> None: