Serialize tensors using int8 views (#16866)

Signed-off-by: Staszek Pasko <staszek@gmail.com>
Co-authored-by: Nick Hill <nhill@redhat.com>

tests/v1/test_serial_utils.py

@@ -47,6 +47,10 @@ def test_encode_decode():
             torch.rand((1, 10), dtype=torch.float32),
             torch.rand((3, 5, 4000), dtype=torch.float64),
             torch.tensor(1984),  # test scalar too
+            # Make sure to test bf16 which numpy doesn't support.
+            torch.rand((3, 5, 1000), dtype=torch.bfloat16),
+            torch.tensor([float("-inf"), float("inf")] * 1024,
+                         dtype=torch.bfloat16),
         ],
         numpy_array=np.arange(512),
         unrecognized=UnrecognizedType(33),
@@ -64,7 +68,7 @@ def test_encode_decode():
     # There should be the main buffer + 4 large tensor buffers
     # + 1 large numpy array. "large" is <= 512 bytes.
     # The two small tensors are encoded inline.
-    assert len(encoded) == 6
+    assert len(encoded) == 8
 
     decoded: MyType = decoder.decode(encoded)
 
@@ -76,7 +80,7 @@ def test_encode_decode():
 
     encoded2 = encoder.encode_into(obj, preallocated)
 
-    assert len(encoded2) == 6
+    assert len(encoded2) == 8
     assert encoded2[0] is preallocated
 
     decoded2: MyType = decoder.decode(encoded2)
@@ -114,15 +118,15 @@ def test_multimodal_kwargs():
 
     total_len = sum(memoryview(x).cast("B").nbytes for x in encoded)
 
-    # expected total encoding length, should be 44536, +-20 for minor changes
+    # expected total encoding length, should be 44559, +-20 for minor changes
-    assert total_len >= 44516 and total_len <= 44556
+    assert total_len >= 44539 and total_len <= 44579
     decoded: MultiModalKwargs = decoder.decode(encoded).mm[0]
     assert all(nested_equal(d[k], decoded[k]) for k in d)
 
 
 def test_multimodal_items_by_modality():
-    e1 = MultiModalFieldElem("audio", "a0", torch.zeros(1000,
+    e1 = MultiModalFieldElem("audio", "a0",
-                                                        dtype=torch.int16),
+                             torch.zeros(1000, dtype=torch.bfloat16),
                              MultiModalBatchedField())
     e2 = MultiModalFieldElem(
         "video",

vllm/v1/serial_utils.py

@@ -80,7 +80,7 @@ class MsgpackEncoder:
 
     def enc_hook(self, obj: Any) -> Any:
         if isinstance(obj, torch.Tensor):
-            return self._encode_ndarray(obj.numpy())
+            return self._encode_tensor(obj)
 
         # Fall back to pickle for object or void kind ndarrays.
         if isinstance(obj, np.ndarray) and obj.dtype.kind not in ('O', 'V'):
@@ -133,9 +133,27 @@ class MsgpackEncoder:
         # backing buffers that we've stashed in `aux_buffers`.
         return obj.dtype.str, obj.shape, data
 
+    def _encode_tensor(
+        self, obj: torch.Tensor
+    ) -> tuple[str, tuple[int, ...], Union[int, memoryview]]:
+        assert self.aux_buffers is not None
+        # this creates a copy of the tensor if it's not already contiguous
+        obj = obj.contiguous()
+        #  view the tensor as a 1D array of bytes
+        arr = obj.view((obj.numel(), )).view(torch.uint8).numpy()
+        if obj.nbytes < self.size_threshold:
+            # Smaller tensors are encoded inline, just like ndarrays.
+            data = msgpack.Ext(CUSTOM_TYPE_RAW_VIEW, arr.data)
+        else:
+            # Otherwise encode index of backing buffer to avoid copy.
+            data = len(self.aux_buffers)
+            self.aux_buffers.append(arr.data)
+        dtype = str(obj.dtype)[6:]  # remove 'torch.' prefix
+        return dtype, obj.shape, data
+
     def _encode_nested_tensors(self, nt: NestedTensors) -> Any:
         if isinstance(nt, torch.Tensor):
-            return self._encode_ndarray(nt.numpy())
+            return self._encode_tensor(nt)
         if isinstance(nt, (int, float)):
             # Although it violates NestedTensors type, MultiModalKwargs
             # values are sometimes floats.
@@ -186,7 +204,7 @@ class MsgpackDecoder:
             if issubclass(t, np.ndarray):
                 return self._decode_ndarray(obj)
             if issubclass(t, torch.Tensor):
-                return torch.from_numpy(self._decode_ndarray(obj))
+                return self._decode_tensor(obj)
             if issubclass(t, MultiModalKwargs):
                 if isinstance(obj, list):
                     return MultiModalKwargs.from_items(
@@ -199,11 +217,24 @@ class MsgpackDecoder:
 
     def _decode_ndarray(self, arr: Any) -> np.ndarray:
         dtype, shape, data = arr
-        # Copy from inline representation, otherwise Torch is unhappy since
+        # zero-copy decode. We assume the ndarray will not be kept around,
-        # the returned memory is non-writeable.
+        # as it now locks the whole received message buffer in memory.
+        buffer = self.aux_buffers[data] if isinstance(data, int) else data
+        return np.ndarray(buffer=buffer, dtype=np.dtype(dtype), shape=shape)
+
+    def _decode_tensor(self, arr: Any) -> torch.Tensor:
+        dtype, shape, data = arr
+        # Copy from inline representation, to decouple the memory storage
+        # of the message from the original buffer. And also make Torch
+        # not complain about a readonly memoryview.
         buffer = self.aux_buffers[data] if isinstance(data, int) \
             else bytearray(data)
-        return np.ndarray(buffer=buffer, dtype=np.dtype(dtype), shape=shape)
+        # Create numpy wrapper around the bytes
+        arr = np.ndarray(buffer=buffer, dtype=np.uint8, shape=(len(buffer), ))
+        torch_dtype = getattr(torch, dtype)
+        assert isinstance(torch_dtype, torch.dtype)
+        # Convert back to proper shape & type
+        return torch.from_numpy(arr).view(torch_dtype).view(shape)
 
     def _decode_mm_items(self, obj: list) -> list[MultiModalKwargsItem]:
         decoded_items = []
@@ -228,7 +259,7 @@ class MsgpackDecoder:
         if not isinstance(obj, list):
             raise TypeError(f"Unexpected NestedTensors contents: {type(obj)}")
         if obj and isinstance(obj[0], str):
-            return torch.from_numpy(self._decode_ndarray(obj))
+            return self._decode_tensor(obj)
         return [self._decode_nested_tensors(x) for x in obj]
 
     def ext_hook(self, code: int, data: memoryview) -> Any: