handle context/multimode mask input smarter

dominoes/train.py

@@ -52,27 +52,27 @@ def train(nets, optimizers, dataset, **parameters):
         # get input for batch
         input = batch["input"]
 
-        # add context inputs for batch if requested (use *context_inputs for consistent handling)
-        context_inputs = []
-        if "context" in batch:
-            context_inputs.append(batch["context"])
-        if "multimode" in batch:
-            context_inputs.append(batch["multimode"])
-
         # get current max output for batch
         max_output = batch.get("max_output", max_possible_output)
 
         # get kwargs for forward pass
         net_kwargs = dict(
             mask=batch.get("mask", None),
-            context_mask=batch.get("context_mask", None),
-            mm_mask=batch.get("mm_mask", None),
             init=batch.get("init", None),
             temperature=temperature,
             thompson=thompson,
             max_output=max_output,
         )
 
+        # add context inputs for batch if requested (use *context_inputs for consistent handling)
+        context_inputs = []
+        if "context" in batch:
+            context_inputs.append(batch["context"])
+            net_kwargs["context_mask"] = batch.get("context_mask", None)
+        if "multimode" in batch:
+            context_inputs.append(batch["multimode"])
+            net_kwargs["mm_mask"] = batch.get("mm_mask", None)
+
         # zero gradients
         for opt in optimizers:
             opt.zero_grad()