[Performance] Batch Send from Tokenizer Manager. (#9436)

python/sglang/srt/managers/io_struct.py

@@ -533,6 +533,21 @@ class TokenizedGenerateReqInput:
     dp_balance_id: int = -1
 
 
+@dataclass
+class BatchTokenizedGenerateReqInput:
+    # The batch of tokenized requests
+    batch: List[TokenizedGenerateReqInput]
+
+    def __len__(self):
+        return len(self.batch)
+
+    def __getitem__(self, i):
+        return self.batch[i]
+
+    def __iter__(self):
+        return iter(self.batch)
+
+
 @dataclass
 class EmbeddingReqInput:
     # The input prompt. It can be a single prompt or a batch of prompts.
@@ -668,6 +683,21 @@ class TokenizedEmbeddingReqInput:
     dp_balance_id: int = -1
 
 
+@dataclass
+class BatchTokenizedEmbeddingReqInput:
+    # The batch of tokenized embedding requests
+    batch: List[TokenizedEmbeddingReqInput]
+
+    def __len__(self):
+        return len(self.batch)
+
+    def __getitem__(self, i):
+        return self.batch[i]
+
+    def __iter__(self):
+        return iter(self.batch)
+
+
 @dataclass
 class BatchTokenIDOut:
     # The request id