diff --git a/nemo/collections/llm/gpt/model/llama.py b/nemo/collections/llm/gpt/model/llama.py
index 804d3892d0c7..04540294d82a 100644
--- a/nemo/collections/llm/gpt/model/llama.py
+++ b/nemo/collections/llm/gpt/model/llama.py
@@ -344,7 +344,10 @@ def apply(self, output_path: Path) -> Path:
 
         target = target.cpu()
         target.save_pretrained(output_path)
-        self.tokenizer.save_pretrained(output_path)
+        try:
+            self.tokenizer.save_pretrained(output_path)
+        except Exception:
+            logging.warning("Failed to save tokenizer")
 
         return output_path
 
diff --git a/nemo/lightning/io/connector.py b/nemo/lightning/io/connector.py
index a38be6ee8f0a..62d5a7e318ca 100644
--- a/nemo/lightning/io/connector.py
+++ b/nemo/lightning/io/connector.py
@@ -226,7 +226,7 @@ def nemo_load(
         from nemo.lightning import MegatronStrategy, Trainer, _strategy_lib
         from nemo.lightning.io.api import load_context
 
-        model = load_context(path).model
+        model = load_context(path, subpath="model")
         _trainer = trainer or Trainer(
             devices=1,
             accelerator="cpu" if cpu else "gpu",