diff --git a/nemo/collections/llm/quantization/quantizer.py b/nemo/collections/llm/quantization/quantizer.py
index 15367cb25aba..2f3e0e1e986e 100644
--- a/nemo/collections/llm/quantization/quantizer.py
+++ b/nemo/collections/llm/quantization/quantizer.py
@@ -13,6 +13,7 @@
 # limitations under the License.
 
 import os
+import shutil
 from dataclasses import dataclass
 from typing import Optional, Union
 
@@ -22,6 +23,7 @@
 from tqdm import tqdm
 
 from nemo.collections import llm
+from nemo.lightning.ckpt_utils import CONTEXT_PATH
 from nemo.utils import logging
 
 from .utils import get_unwrapped_mcore_model
@@ -259,7 +261,7 @@ def loop(model):
 
         return loop
 
-    def export(self, model: llm.GPTModel) -> None:
+    def export(self, model: llm.GPTModel, model_dir: str) -> None:
         assert self.export_config is not None, "Export config is not set"
         # TODO: Add sample generate
         # TODO: Support megatron_amp_O2
@@ -277,15 +279,16 @@ def export(self, model: llm.GPTModel) -> None:
             use_nfs_workspace=use_nfs_workspace,
         )
 
-        dist.barrier()  # Wait until all ranks complete export_model_config step
-        logging.info(f"Export succeeded, model has been exported to {export_dir}. Saving tokenizer if possible...")
+        # Save the model context in order to restore its tokenizer later. The destination
+        # path is "nemo_context" as this name is used in nemo.export to setup tokenizer.
+        shutil.copytree(
+            os.path.join(model_dir, CONTEXT_PATH),
+            os.path.join(export_dir, "nemo_context"),
+            dirs_exist_ok=True,
+        )
+        logging.info(f"Model context saved.")
 
-        if dist.get_rank() == 0:
-            try:
-                tokenizer_dst = os.path.join(export_dir, 'tokenizer')
-                model.tokenizer.tokenizer.save_pretrained(tokenizer_dst)
-            except Exception as err:
-                logging.warning("Could not save the tokenizer: " + str(err))
+        logging.info(f"Export succeeded, model has been exported to {export_dir}.")
 
 
 def get_calib_data_iter(
diff --git a/nemo/export/tensorrt_llm.py b/nemo/export/tensorrt_llm.py
index fb43224d59a9..08b0b822cad4 100644
--- a/nemo/export/tensorrt_llm.py
+++ b/nemo/export/tensorrt_llm.py
@@ -37,12 +37,12 @@
 from nemo.export.trt_llm.converter.utils import init_model_parallel_from_nemo
 from nemo.export.trt_llm.nemo_ckpt_loader.nemo_file import (
     build_tokenizer,
-    get_tokenzier,
+    get_tokenizer,
     is_nemo_file,
     load_nemo_model,
 )
 from nemo.export.trt_llm.qnemo import qnemo_to_tensorrt_llm
-from nemo.export.trt_llm.qnemo.tokenizer_utils import get_nmt_tokenizer
+from nemo.export.trt_llm.qnemo.tokenizer_utils import TOKENIZER_CONFIG_FILE, get_nmt_tokenizer
 from nemo.export.trt_llm.qnemo.utils import is_qnemo_checkpoint
 from nemo.export.trt_llm.tensorrt_llm_build import build_and_save_engine
 from nemo.export.trt_llm.tensorrt_llm_run import (
@@ -294,7 +294,14 @@ def export(
                 else:
                     unpack_tarball(nemo_checkpoint_path, tmp_dir.name)
                     nemo_checkpoint_path = tmp_dir.name
-                self.tokenizer = get_nmt_tokenizer(nemo_checkpoint_path)
+
+                if os.path.exists(os.path.join(nemo_checkpoint_path, TOKENIZER_CONFIG_FILE)):
+                    # Instantiate tokenizer for a legacy "Nemo 1" quantized checkpoint from a tokenizer config.
+                    # Note that using the config is deprecated and it will be removed in future releases.
+                    LOGGER.warning("Detected legacy tokenizer_config.yaml, using it to build tokenizer.")
+                    self.tokenizer = get_nmt_tokenizer(nemo_checkpoint_path)
+                else:
+                    self.tokenizer = get_tokenizer(nemo_checkpoint_path)
 
                 qnemo_to_tensorrt_llm(
                     nemo_checkpoint_path=nemo_checkpoint_path,
@@ -1092,7 +1099,7 @@ def _load(self):
             if len(folders) > 0:
                 try:
                     self._load_config_file()
-                    self.tokenizer = get_tokenzier(Path(os.path.join(self.model_dir)))
+                    self.tokenizer = get_tokenizer(self.model_dir)
                     self.model = load(
                         tokenizer=self.tokenizer,
                         engine_dir=self.model_dir,
diff --git a/nemo/export/trt_llm/nemo_ckpt_loader/nemo_file.py b/nemo/export/trt_llm/nemo_ckpt_loader/nemo_file.py
index 407a7ce600c9..23d227d32acf 100644
--- a/nemo/export/trt_llm/nemo_ckpt_loader/nemo_file.py
+++ b/nemo/export/trt_llm/nemo_ckpt_loader/nemo_file.py
@@ -283,16 +283,17 @@ def copy_tokenizer_files(config, out_dir):
                 outfile.write(infile.read())
 
 
-def get_tokenzier(tokenizer_dir_or_path: Path) -> PreTrainedTokenizer:
-    """Loads the tokenizer from the decoded NEMO weights dir."""
+def get_tokenizer(tokenizer_dir_or_path: Union[str, Path]) -> PreTrainedTokenizer:
+    """Loads the tokenizer from the decoded NeMo weights dir."""
+    tokenizer_dir_or_path = Path(tokenizer_dir_or_path)
     if (tokenizer_dir_or_path / "nemo_context").exists():
         from nemo.lightning import io
 
         tokenizer_spec = io.load_context((tokenizer_dir_or_path / "nemo_context"), subpath="model.tokenizer")
         return build_tokenizer(tokenizer_spec)
     else:
-        if os.path.isdir(os.path.join(tokenizer_dir_or_path, "huggingface_tokenizer")):
-            return AutoTokenizer.from_pretrained(os.path.join(tokenizer_dir_or_path, "huggingface_tokenizer"))
+        if (tokenizer_dir_or_path / "huggingface_tokenizer").is_dir():
+            return AutoTokenizer.from_pretrained(tokenizer_dir_or_path / "huggingface_tokenizer")
 
         model_path = (
             tokenizer_dir_or_path / "tokenizer.model" if tokenizer_dir_or_path.is_dir() else tokenizer_dir_or_path
diff --git a/nemo/export/trt_llm/qnemo/tokenizer_utils.py b/nemo/export/trt_llm/qnemo/tokenizer_utils.py
index 36efa9259f9d..beca40bcd3d7 100644
--- a/nemo/export/trt_llm/qnemo/tokenizer_utils.py
+++ b/nemo/export/trt_llm/qnemo/tokenizer_utils.py
@@ -29,11 +29,6 @@
 def get_nmt_tokenizer(nemo_checkpoint_path: str):
     """Build tokenizer from Nemo tokenizer config."""
 
-    tokenizer_dir = os.path.join(nemo_checkpoint_path, TOKENIZER_DIR)
-    if os.path.exists(tokenizer_dir):
-        print(f"Initializing tokenizer from {TOKENIZER_DIR} directory")
-        return AutoTokenizer.from_pretrained(tokenizer_dir)
-
     print(f"Initializing tokenizer from {TOKENIZER_CONFIG_FILE}")
     tokenizer_cfg = OmegaConf.load(os.path.join(nemo_checkpoint_path, TOKENIZER_CONFIG_FILE))
 
diff --git a/nemo/utils/model_utils.py b/nemo/utils/model_utils.py
index b417c088b22e..5d7d019c6099 100644
--- a/nemo/utils/model_utils.py
+++ b/nemo/utils/model_utils.py
@@ -724,6 +724,10 @@ def save_artifacts(model, output_dir: str, use_abspath: bool = False) -> None:
     app_state = AppState()
     model_file = app_state.model_restore_path
     model_cfg = copy.deepcopy(model.cfg)
+
+    if model_cfg.tokenizer.library == "huggingface":
+        model.tokenizer.save_pretrained(os.path.join(output_dir, "huggingface_tokenizer"))
+
     if not hasattr(model, "artifacts"):
         if hasattr(model_cfg, "tokenizer"):
             OmegaConf.save(model_cfg.tokenizer, os.path.join(output_dir, "tokenizer_config.yaml"))
diff --git a/scripts/llm/ptq.py b/scripts/llm/ptq.py
index 0fd2c5682e8a..c04d32290e5f 100644
--- a/scripts/llm/ptq.py
+++ b/scripts/llm/ptq.py
@@ -92,7 +92,7 @@ def main():
     quantizer = quantization.Quantizer(quantization_config, export_config)
     model = quantization.load_with_modelopt_layer_spec(args.nemo_checkpoint, args.calib_tp, args.calib_pp)
     model = quantizer.quantize(model)
-    quantizer.export(model)
+    quantizer.export(model, args.nemo_checkpoint)
 
 
 if __name__ == '__main__':