From 3d42b7e7b02517391799e26de0bf790bc914460c Mon Sep 17 00:00:00 2001
From: Yuhao Yang <47235274+yhyang201@users.noreply.github.com>
Date: Thu, 18 Dec 2025 12:28:06 +0800
Subject: [PATCH] unified management of environment variables for vlm cuda ipc
 transport  (#14501)

---
 python/sglang/srt/environ.py                         |  5 +++++
 .../srt/multimodal/processors/base_processor.py      | 12 +++---------
 python/sglang/srt/utils/cuda_ipc_transport_utils.py  | 12 +++---------
 3 files changed, 11 insertions(+), 18 deletions(-)

diff --git a/python/sglang/srt/environ.py b/python/sglang/srt/environ.py
index db7f667dd..417b9d378 100644
--- a/python/sglang/srt/environ.py
+++ b/python/sglang/srt/environ.py
@@ -330,6 +330,11 @@ class Envs:
     SGLANG_RESIZE_RESAMPLE = EnvStr("")
     SGLANG_MM_BUFFER_SIZE_MB = EnvInt(0)
 
+    # VLM Item CUDA IPC Transport
+    SGLANG_USE_CUDA_IPC_TRANSPORT=EnvBool(False)
+    SGLANG_MM_FEATURE_CACHE_MB = EnvInt(4 * 1024)
+    SGLANG_MM_ITEM_MEM_POOL_RECYCLE_INTERVAL_SEC = EnvFloat(0.05)
+
     # Release & Resume Memory
     SGLANG_MEMORY_SAVER_CUDA_GRAPH = EnvBool(False)
 
diff --git a/python/sglang/srt/multimodal/processors/base_processor.py b/python/sglang/srt/multimodal/processors/base_processor.py
index 6553178e6..3eeabefb1 100644
--- a/python/sglang/srt/multimodal/processors/base_processor.py
+++ b/python/sglang/srt/multimodal/processors/base_processor.py
@@ -12,15 +12,9 @@ import torch
 from PIL import Image
 from transformers import BaseImageProcessorFast
 
+from sglang.srt.environ import envs
 from sglang.srt.managers.schedule_batch import Modality, MultimodalDataItem
-from sglang.srt.utils import (
-    get_bool_env_var,
-    is_npu,
-    load_audio,
-    load_image,
-    load_video,
-    logger,
-)
+from sglang.srt.utils import is_npu, load_audio, load_image, load_video, logger
 from sglang.srt.utils.cuda_ipc_transport_utils import (
     MM_FEATURE_CACHE_SIZE,
     MM_ITEM_MEMORY_POOL_RECYCLE_INTERVAL,
@@ -30,7 +24,7 @@ from sglang.srt.utils.cuda_ipc_transport_utils import (
 
 _is_npu = is_npu()
 
-SGL_USE_CUDA_IPC = get_bool_env_var("SGLANG_USE_CUDA_IPC_TRANSPORT")
+SGL_USE_CUDA_IPC = envs.SGLANG_USE_CUDA_IPC_TRANSPORT.get()
 
 
 @dataclasses.dataclass
diff --git a/python/sglang/srt/utils/cuda_ipc_transport_utils.py b/python/sglang/srt/utils/cuda_ipc_transport_utils.py
index 268bb57a9..6d76242ae 100644
--- a/python/sglang/srt/utils/cuda_ipc_transport_utils.py
+++ b/python/sglang/srt/utils/cuda_ipc_transport_utils.py
@@ -8,21 +8,15 @@ from typing import Tuple
 import numpy as np
 import torch
 
+from sglang.srt.environ import envs
 from sglang.srt.server_args import get_global_server_args
-from sglang.srt.utils import get_float_env_var, get_int_env_var
 
 logger = logging.getLogger(__name__)
 
-MM_FEATURE_CACHE_SIZE = (
-    4 * 1024 * 1024 * 1024
-    if not get_int_env_var("SGLANG_MM_FEATURE_CACHE_MB")
-    else get_int_env_var("SGLANG_MM_FEATURE_CACHE_MB") * 1024 * 1024
-)
+MM_FEATURE_CACHE_SIZE = envs.SGLANG_MM_FEATURE_CACHE_MB.get() * 1024 * 1024
 
 MM_ITEM_MEMORY_POOL_RECYCLE_INTERVAL = (
-    0.05
-    if not get_float_env_var("SGLANG_MM_ITEM_MEM_POOL_RECYCLE_INTERVAL_SEC")
-    else get_float_env_var("SGLANG_MM_ITEM_MEM_POOL_RECYCLE_INTERVAL_SEC")
+    envs.SGLANG_MM_ITEM_MEM_POOL_RECYCLE_INTERVAL_SEC.get()
 )
 
 SHM_LOCK_FILE = "/tmp/shm_wr_lock.lock"