Remove inf value for chunked prefill size (#812)

python/sglang/srt/managers/controller/tp_worker.py

@@ -442,8 +442,11 @@ class ModelTpServer:
                 else:
                     # Add this request to the running batch
                     if (
-                        new_batch_input_tokens + req.extend_input_len
-                        <= self.chunked_prefill_size
+                        self.chunked_prefill_size is None
+                        or (
+                            new_batch_input_tokens + req.extend_input_len
+                            <= self.chunked_prefill_size
+                        )
                         or (
                             req.return_logprob and req.normalized_prompt_logprob is None
                         )

python/sglang/srt/server_args.py

@@ -87,8 +87,6 @@ class ServerArgs:
     node_rank: Optional[int] = None
 
     def __post_init__(self):
-        if self.chunked_prefill_size is None:
-            self.chunked_prefill_size = 1 << 30
         if self.tokenizer_path is None:
             self.tokenizer_path = self.model_path
         if self.mem_fraction_static is None:
@@ -414,7 +412,7 @@ class ServerArgs:
         ), "multi-node data parallel is not supported"
 
         assert not (
-            self.chunked_prefill_size < (1 << 30) and self.disable_radix_cache
+            self.chunked_prefill_size is not None and self.disable_radix_cache
         ), "chunked prefill is not supported with radix cache disabled currently"