[Frontend] Remove custom_cache_manager (#13791)

Signed-off-by: fulvius31 <[email protected]>

Author: fulvius31

vllm/executor/multiproc_worker_utils.py

@@ -16,12 +16,8 @@
 
 from vllm.config import VllmConfig
 from vllm.logger import init_logger
-from vllm.triton_utils.importing import HAS_TRITON
 from vllm.utils import _check_multiproc_method, get_mp_context, run_method
 
-if HAS_TRITON:
-    from vllm.triton_utils import maybe_set_triton_cache_manager
-
 logger = init_logger(__name__)
 
 T = TypeVar('T')
@@ -314,7 +310,3 @@ def set_multiprocessing_worker_envs(parallel_config):
             current_parallelism, default_omp_num_threads)
         os.environ["OMP_NUM_THREADS"] = str(default_omp_num_threads)
         torch.set_num_threads(default_omp_num_threads)
-
-    # workaround for https://github.com/vllm-project/vllm/issues/6103
-    if HAS_TRITON and parallel_config.world_size > 1:
-        maybe_set_triton_cache_manager()

vllm/triton_utils/__init__.py

@@ -2,11 +2,4 @@
 
 from vllm.triton_utils.importing import HAS_TRITON
 
-__all__ = ["HAS_TRITON"]
-
-if HAS_TRITON:
-
-    from vllm.triton_utils.custom_cache_manager import (
-        maybe_set_triton_cache_manager)
-
-    __all__ += ["maybe_set_triton_cache_manager"]
+__all__ = ["HAS_TRITON"]