minor code sync (#12403)

python/sglang/srt/managers/io_struct.py

@@ -1473,6 +1473,16 @@ class WatchLoadUpdateReq(BaseReq):
     loads: List[GetLoadReqOutput]
 
 
+@dataclass
+class SetInjectDumpMetadataReqInput(BaseReq):
+    dump_metadata: Dict[str, Any]
+
+
+@dataclass
+class SetInjectDumpMetadataReqOutput(BaseReq):
+    success: bool
+
+
 @dataclass
 class LazyDumpTensorsReqInput(BaseReq):
     pass
@@ -1504,6 +1514,3 @@ def _check_all_req_types():
             raise ValueError(
                 f"{name} is a subclass of BaseReq but not follow the naming convention."
             )
-
-
-_check_all_req_types()

python/sglang/srt/metrics/collector.py

@@ -870,13 +870,13 @@ class TokenizerMetricsCollector:
     def check_time_to_first_token_straggler(self, value: float) -> bool:
         his = self.histogram_time_to_first_token.labels(**self.labels)
         total_observations = sum(bucket._value for bucket in his._buckets)
-        if total_observations < 1000:
+        if total_observations < 100:
             return False
-        p999_threshold = total_observations * 0.999
+        p99_threshold = total_observations * 0.99
         cumulative_count = 0
         for i, bucket in enumerate(his._buckets):
             cumulative_count += bucket._value
-            if cumulative_count > p999_threshold:
+            if cumulative_count > p99_threshold:
                 return value >= his._upper_bounds[i]
         return False
 

python/sglang/srt/server_args.py

@@ -135,6 +135,8 @@ GRAMMAR_BACKEND_CHOICES = ["xgrammar", "outlines", "llguidance", "none"]
 
 DETERMINISTIC_ATTENTION_BACKEND_CHOICES = ["flashinfer", "fa3", "triton"]
 
+RADIX_SUPPORTED_DETERMINISTIC_ATTENTION_BACKEND = ["fa3", "triton"]
+
 DEFAULT_LORA_EVICTION_POLICY = "lru"
 
 NSA_CHOICES = [
@@ -190,6 +192,10 @@ def add_deterministic_attention_backend_choices(choices):
     DETERMINISTIC_ATTENTION_BACKEND_CHOICES.extend(choices)
 
 
+def add_radix_supported_deterministic_attention_backend_choices(choices):
+    RADIX_SUPPORTED_DETERMINISTIC_ATTENTION_BACKEND.extend(choices)
+
+
 def add_radix_eviction_policy_choices(choices):
     RADIX_EVICTION_POLICY_CHOICES.extend(choices)
 
@@ -1753,13 +1759,17 @@ class ServerArgs:
                     f"but you explicitly specified '{self.attention_backend}'."
                 )
 
-            if self.attention_backend not in ["fa3", "triton"]:
-                if is_deepseek_model:
+            if is_deepseek_model:
+                if self.attention_backend not in ["fa3", "triton"]:
                     raise ValueError(
-                        f"Currently only fa3 and triton attention backends are supported for deterministic inference with DeepSeek models. But you're using {self.attention_backend}."
+                        f"Currently only {RADIX_SUPPORTED_DETERMINISTIC_ATTENTION_BACKEND} attention backends are supported for deterministic inference with DeepSeek models. But you're using {self.attention_backend}."
                     )
 
-                # Currently, only FA3 and Triton supports radix cache. Support for other backends is in progress
+            if (
+                self.attention_backend
+                not in RADIX_SUPPORTED_DETERMINISTIC_ATTENTION_BACKEND
+            ):
+                # Currently, only certain backends support radix cache. Support for other backends is in progress
                 self.disable_radix_cache = True
                 logger.warning(
                     f"Currently radix cache is not compatible with {self.attention_backend} attention backend for deterministic inference. It will be supported in the future."