Convert tensors to bytes instead of numpy in multiprocessing result-q…

…ueue (#20005)

src/lightning/pytorch/strategies/launchers/multiprocessing.py

@@ -11,6 +11,7 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import io
 import logging
 import os
 import queue
@@ -19,7 +20,6 @@
 from dataclasses import dataclass
 from typing import Any, Callable, Dict, List, Literal, NamedTuple, Optional, Union
 
-import numpy as np
 import torch
 import torch.backends.cudnn
 import torch.multiprocessing as mp
@@ -226,7 +226,7 @@ def _collect_rank_zero_results(self, trainer: "pl.Trainer", results: Any) -> Opt
 
     def get_extra_results(self, trainer: "pl.Trainer") -> Dict[str, Any]:
         """Gather extra state from the Trainer and return it as a dictionary for sending back to the main process. To
-        avoid issues with memory sharing, we cast the data to numpy.
+        avoid issues with memory sharing, we convert tensors to bytes.
 
         Args:
             trainer: reference to the Trainer.
@@ -236,14 +236,15 @@ def get_extra_results(self, trainer: "pl.Trainer") -> Dict[str, Any]:
             process this output.
 
         """
-        callback_metrics: dict = apply_to_collection(
-            trainer.callback_metrics, Tensor, lambda x: x.cpu().numpy()
-        )  # send as numpy to avoid issues with memory sharing
-        return {"callback_metrics": callback_metrics}
+        callback_metrics = apply_to_collection(trainer.callback_metrics, Tensor, lambda t: t.cpu())
+        buffer = io.BytesIO()
+        torch.save(callback_metrics, buffer)
+        # send tensors as bytes to avoid issues with memory sharing
+        return {"callback_metrics_bytes": buffer.getvalue()}
 
     def update_main_process_results(self, trainer: "pl.Trainer", extra: Dict[str, Any]) -> None:
         """Retrieve the :attr:`trainer.callback_metrics` dictionary from the given queue. To preserve consistency, we
-        cast back the data to ``torch.Tensor``.
+        convert bytes back to ``torch.Tensor``.
 
         Args:
             trainer: reference to the Trainer.
@@ -252,8 +253,9 @@ def update_main_process_results(self, trainer: "pl.Trainer", extra: Dict[str, An
 
         """
         # NOTE: `get_extra_results` needs to be called before
-        callback_metrics = extra["callback_metrics"]
-        trainer.callback_metrics.update(apply_to_collection(callback_metrics, np.ndarray, lambda x: torch.tensor(x)))
+        callback_metrics_bytes = extra["callback_metrics_bytes"]
+        callback_metrics = torch.load(io.BytesIO(callback_metrics_bytes))
+        trainer.callback_metrics.update(callback_metrics)
 
     @override
     def kill(self, signum: _SIGNUM) -> None: