update examples from master

huggingface · Nov 17, 2018 · 47a7d4e · 47a7d4e
2 parents c8cba67 + 02173a1
commit 47a7d4e
Show file tree

Hide file tree

Showing 3 changed files with 11 additions and 6 deletions.
diff --git a/README.md b/README.md
@@ -395,7 +395,7 @@ python run_squad.py \
   --num_train_epochs 2.0 \
   --max_seq_length 384 \
   --doc_stride 128 \
-  --output_dir ../debug_squad/
+  --output_dir /tmp/debug_squad/
 ```
 
 Training with the previous hyper-parameters gave us the following results:

diff --git a/examples/run_classifier.py b/examples/run_classifier.py
@@ -327,11 +327,14 @@ def set_optimizer_params_grad(named_params_optimizer, named_params_model, test_n
         if name_opti != name_model:
             logger.error("name_opti != name_model: {} {}".format(name_opti, name_model))
             raise ValueError
-        if test_nan and torch.isnan(param_model.grad).sum() > 0:
-            is_nan = True
-        if param_opti.grad is None:
-            param_opti.grad = torch.nn.Parameter(param_opti.data.new().resize_(*param_opti.data.size()))
-        param_opti.grad.data.copy_(param_model.grad.data)
+        if param_model.grad is not None:
+            if test_nan and torch.isnan(param_model.grad).sum() > 0:
+                is_nan = True
+            if param_opti.grad is None:
+                param_opti.grad = torch.nn.Parameter(param_opti.data.new().resize_(*param_opti.data.size()))
+            param_opti.grad.data.copy_(param_model.grad.data)
+        else:
+            param_opti.grad = None
     return is_nan
 
 def main():

diff --git a/examples/run_squad.py b/examples/run_squad.py
@@ -693,6 +693,8 @@ def set_optimizer_params_grad(named_params_optimizer, named_params_model, test_n
             if param_opti.grad is None:
                 param_opti.grad = torch.nn.Parameter(param_opti.data.new().resize_(*param_opti.data.size()))
             param_opti.grad.data.copy_(param_model.grad.data)
+        else:
+            param_opti.grad = None
     return is_nan
 
 def main():