[feat] support minimum token load balance in dp attention (#7379)

python/sglang/srt/managers/io_struct.py

@@ -523,6 +523,9 @@ class TokenizedGenerateReqInput:
     # For data parallel rank routing
     data_parallel_rank: Optional[int] = None
 
+    # For dp balance
+    dp_balance_id: int = -1
+
 
 @dataclass
 class EmbeddingReqInput:
@@ -648,6 +651,8 @@ class TokenizedEmbeddingReqInput:
     token_type_ids: List[int]
     # Dummy sampling params for compatibility
     sampling_params: SamplingParams
+    # For dp balance
+    dp_balance_id: int = -1
 
 
 @dataclass