add custom loss

Author: NihalHarish

smdebug/pytorch/hook.py

@@ -3,6 +3,7 @@
 # Third Party
 import torch
 import torch.distributed as dist
+from torch.nn.modules.loss import _Loss
 
 # First Party
 from smdebug.core.collection import DEFAULT_PYTORCH_COLLECTIONS, CollectionKeys
@@ -154,6 +155,9 @@ def forward_hook(self, module, inputs, outputs):
         if not self._get_collections_to_save_for_step():
             return
 
+        if isinstance(module, _Loss):
+            module._module_name = module._get_name()
+
         module_name = module._module_name
         # This overwhelms the logs; turn back on if you really need it
         # logger.debug("Processing the global step {0} for module {1}".format(self.step, module_name))

tests/zero_code_change/test_pytorch_integration.py

@@ -23,20 +23,43 @@
 from smdebug.core.utils import SagemakerSimulator, ScriptSimulator
 
 
+class CustomCrossEntropyLoss(nn.modules.loss._WeightedLoss):
+    __constants__ = ["weight", "ignore_index", "reduction"]
+
+    def __init__(
+        self, weight=None, size_average=None, ignore_index=-100, reduce=None, reduction="mean"
+    ):
+        super(CustomCrossEntropyLoss, self).__init__(weight, size_average, reduce, reduction)
+        self.ignore_index = ignore_index
+
+    def forward(self, input, target):
+        return F.cross_entropy(
+            input,
+            target,
+            weight=self.weight,
+            ignore_index=self.ignore_index,
+            reduction=self.reduction,
+        )
+
+
 @pytest.mark.skipif(
     torch.__version__ == "1.7.0",
     reason="Disabling the test temporarily until we root cause the version incompatibility",
 )
 @pytest.mark.parametrize("script_mode", [False])
 @pytest.mark.parametrize("use_loss_module", [True, False])
-def test_pytorch(script_mode, use_loss_module):
+@pytest.mark.parametrize("use_custom_loss_module", [True, False])
+def test_pytorch(script_mode, use_loss_module, use_custom_loss_module):
     smd.del_hook()
 
     sim_class = ScriptSimulator if script_mode else SagemakerSimulator
     with sim_class() as sim:
         trainloader, testloader = get_dataloaders()
         net = Net()
-        criterion = nn.CrossEntropyLoss()
+        if use_custom_loss_module:
+            criterion = CustomCrossEntropyLoss()
+        else:
+            criterion = nn.CrossEntropyLoss()
         optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
 
         if script_mode: