vlm: support SGLANG_MM_SKIP_COMPUTE_HASH for bypassing multimodal feature hashing (#16555)

python/sglang/srt/environ.py

@@ -377,6 +377,8 @@ class Envs:
     SGLANG_MM_BUFFER_SIZE_MB = EnvInt(0)
     SGLANG_MM_PRECOMPUTE_HASH = EnvBool(False)
     SGLANG_VIT_ENABLE_CUDA_GRAPH = EnvBool(False)
+    SGLANG_MM_SKIP_COMPUTE_HASH = EnvBool(False)
+
 
     # VLM Item CUDA IPC Transport
     SGLANG_USE_CUDA_IPC_TRANSPORT = EnvBool(False)

python/sglang/srt/managers/schedule_batch.py

@@ -258,6 +258,12 @@ class MultimodalDataItem:
 
         from sglang.srt.managers.mm_utils import hash_feature
 
+        if envs.SGLANG_MM_SKIP_COMPUTE_HASH.get():
+            import uuid
+
+            self.hash = uuid.uuid4().int
+            self.pad_value = self.hash % (1 << 30)
+            return
         if self.hash is None:
             if self.feature is not None:
                 hashed_feature = self.feature