Import the MTP block's final_layernorm (repeated-MTP path)

The repeated-MTP import branch loops over the inner MTP layers but never
imports the MTP block's own final_layernorm, leaving it random-initialized.
Add the import after the loop, keyed to the last inner layer index
(layer_id - 1) so it round-trips with the export side, which writes
mtp.layers.{last}.final_layernorm.

Bug found by @jenchen13.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: yeyu-nvidia

modelopt/torch/export/plugins/megatron_importer.py

@@ -860,6 +860,11 @@ def _import_state_dict(self):
                         )
 
                     layer_id += 1
+
+                # Import the MTP block's own final_layernorm into the last inner layer
+                # index; without this it is left random-initialized. Mirrors the export
+                # side, which writes mtp.layers.{last}.final_layernorm.
+                self.rules["mtp.final_layernorm"](mtp.final_layernorm, layer_id - 1)
             else:  # non-repeated MTP
                 # MTP is the last layer in DeepSeek V3/R1
                 layer_id += 1