Improved comments

Author: shino16

thunder/dynamo/compiler.py

@@ -135,7 +135,8 @@ def __call__(self, gm: torch.fx.GraphModule, sample_args: list[torch.SymInt, tor
 
         remove_empty_autocast(gm)
 
-        # Convert tag_activation_checkpoint operators, which is merely a tagger for torch.compile stack, to actual checkpoint calls
+        # Convert tag_activation_checkpoint operators, which is meaningless in eager mode, to actual checkpoint calls
+        # This will not be needed when we have found a way to make tag_activation_checkpoint fall back to PyTorch's backend
         convert_checkpoint_tags(gm)
 
         # The whole graph may not be supported by `thunder`, so we split it in `thunder` supported sections

thunder/dynamo/utils.py

@@ -635,7 +635,7 @@ def convert_checkpoint_tags(gm: torch.fx.GraphModule):
     """
     Replaces tag_activation_checkpoint operators with torch.utils.checkpoint.checkpoint calls.
 
-    tag_activation_checkpoint only marks nodes for recomputation within torch.compile but does not execute checkpointing itself.
+    tag_activation_checkpoint only marks nodes for torch.compile stack but does not execute actual checkpointing in eager mode.
     """
     for n in gm.graph.nodes:
         if n.op == "call_function" and n.target is torch.ops.higher_order.tag_activation_checkpoint: