add ability to return logits in eval

Author: sdtblck

deepspeed/runtime/pipe/engine.py

@@ -65,6 +65,8 @@ def __init__(self, *super_args, **super_kwargs):
 
         # We schedule the all-reduces, so disable it in super().backward()
         self.enable_backward_allreduce = False
+        self.eval_return_logits = False
+        self.outputs = None
 
         # used to disable the pipeline all-reduce when used with 1-bit Adam/1-bit LAMB
         self.pipeline_enable_backward_allreduce = True
@@ -336,7 +338,7 @@ def train_batch(self, data_iter=None):
         # TODO: should return precisely what loss returned and allow others to be queried?
         return self.agg_train_loss
 
-    def eval_batch(self, data_iter):
+    def eval_batch(self, data_iter, return_logits=False):
         """Evaluate the pipeline on a batch of data from ``data_iter``. The
         engine will evaluate ``self.train_batch_size()`` total samples
         collectively across all workers.
@@ -363,7 +365,7 @@ def eval_batch(self, data_iter):
         Returns:
             The arithmetic mean of the losses computed this batch.
         """
-
+        self.eval_return_logits = return_logits
         self.module.eval()
         self.total_loss = None
 
@@ -393,7 +395,11 @@ def eval_batch(self, data_iter):
 
         # Reset any buffers that may have been populated during the forward passes.
         # ds_checkpointing.reset()
-
+        self.eval_return_logits = False
+        if return_logits:
+            outputs = self.outputs
+            self.outputs = None
+            return self.agg_eval_loss, outputs
         return self.agg_eval_loss
 
     def inference_batch(self, data_iter):
@@ -666,6 +672,8 @@ def _exec_forward_pass(self, buffer_id):
             else:
                 # Some models just return loss from forward()
                 self.loss = outputs
+            if self.eval_return_logits:
+                self.outputs = outputs
 
             if isinstance(self.loss, torch.Tensor):
                 if self.total_loss is None: