Revert "Deprecate global_server_args_dict" (#11520)

python/sglang/srt/sampling/sampling_batch_info.py

@@ -2,6 +2,7 @@ from __future__ import annotations
 
 import dataclasses
 import logging
+import threading
 from typing import TYPE_CHECKING, Any, Callable, Dict, List, Optional, Tuple
 
 import torch
@@ -9,7 +10,6 @@ import torch
 import sglang.srt.sampling.penaltylib as penaltylib
 from sglang.srt.sampling.custom_logit_processor import CustomLogitProcessor
 from sglang.srt.sampling.sampling_params import TOP_K_ALL
-from sglang.srt.server_args import get_global_server_args
 
 if TYPE_CHECKING:
     from sglang.srt.managers.schedule_batch import ScheduleBatch
@@ -66,10 +66,16 @@ class SamplingBatchInfo:
     # Handle logit bias
     logit_bias: Optional[torch.Tensor] = None
 
+    @classmethod
+    def _get_global_server_args_dict(cls):
+        from sglang.srt.managers.schedule_batch import global_server_args_dict
+
+        return global_server_args_dict
+
     @classmethod
     def from_schedule_batch(cls, batch: ScheduleBatch, vocab_size: int):
-        global_server_args = get_global_server_args()
-        enable_deterministic = global_server_args.enable_deterministic_inference
+        global_server_args_dict = cls._get_global_server_args_dict()
+        enable_deterministic = global_server_args_dict["enable_deterministic_inference"]
 
         reqs = batch.reqs
         device = batch.device
@@ -106,9 +112,10 @@ class SamplingBatchInfo:
                         logit_bias[i, int(key)] = value
 
         # Check if any request has custom logit processor
-        has_custom_logit_processor = (
-            global_server_args.enable_custom_logit_processor
-            and any(r.custom_logit_processor for r in reqs)  # check the flag first.
+        has_custom_logit_processor = global_server_args_dict[
+            "enable_custom_logit_processor"
+        ] and any(  # check the flag first.
+            r.custom_logit_processor for r in reqs
         )  # then check the requests.
 
         if has_custom_logit_processor: