[small][batch invariance] Rename the env and internal flags to simplify usage (#26855)

Signed-off-by: Bram Wasti <bwasti@meta.com>

vllm/distributed/device_communicators/all_reduce_utils.py

@@ -20,7 +20,7 @@ import vllm.envs as envs
 from vllm.distributed.device_communicators.cuda_wrapper import CudaRTLibrary
 from vllm.logger import init_logger
 from vllm.model_executor.layers.batch_invariant import (
-    vllm_kernel_override_batch_invariant,
+    vllm_is_batch_invariant,
 )
 from vllm.utils import cuda_device_count_stateless, update_environment_variables
 
@@ -74,7 +74,7 @@ def should_nccl_symm_mem_allreduce(world_size: int, input_tensor: torch.Tensor)
         is_symmetric_memory_enabled,
     )
 
-    if vllm_kernel_override_batch_invariant():
+    if vllm_is_batch_invariant():
         return False
 
     if not is_symmetric_memory_enabled():

vllm/distributed/device_communicators/symm_mem.py

@@ -10,7 +10,7 @@ from vllm.distributed.device_communicators.all_reduce_utils import (
 )
 from vllm.logger import init_logger
 from vllm.model_executor.layers.batch_invariant import (
-    vllm_kernel_override_batch_invariant,
+    vllm_is_batch_invariant,
 )
 from vllm.platforms import current_platform
 
@@ -103,7 +103,7 @@ class SymmMemCommunicator:
             return
         self.force_multimem = force_multimem
         self.disabled = False
-        if vllm_kernel_override_batch_invariant():
+        if vllm_is_batch_invariant():
             self.disabled = True
 
     def should_use_symm_mem(self, inp: torch.Tensor):