[Misc] Various cleanups for MM input processing (#29970)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>

vllm/utils/serial_utils.py

@@ -1,6 +1,7 @@
 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
 import base64
+import io
 import sys
 from dataclasses import dataclass
 from typing import Literal
@@ -52,6 +53,15 @@ Endianness = Literal["native", "big", "little"]
 EncodingFormat = Literal["float", "base64", "bytes"]
 
 
+def tensor2base64(x: torch.Tensor) -> str:
+    with io.BytesIO() as buf:
+        torch.save(x, buf)
+        buf.seek(0)
+        binary_data = buf.read()
+
+    return base64.b64encode(binary_data).decode("utf-8")
+
+
 def tensor2binary(
     tensor: torch.Tensor, embed_dtype: EmbedDType, endianness: Endianness
 ) -> bytes: