diff --git a/.flake8 b/.flake8
index 6c31553..998689f 100644
--- a/.flake8
+++ b/.flake8
@@ -5,5 +5,3 @@ per-file-ignores =
     src/dom_tokenizers/**/__init__.py: F401
     # line too long
     src/dom_tokenizers/pre_tokenizers/dom_snapshot.py: E501
-    # module level import not at top of file
-    src/dom_tokenizers/train.py: E402
diff --git a/src/dom_tokenizers/internal/__init__.py b/src/dom_tokenizers/internal/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/src/dom_tokenizers/internal/transformers.py b/src/dom_tokenizers/internal/transformers.py
new file mode 100644
index 0000000..3ddc75c
--- /dev/null
+++ b/src/dom_tokenizers/internal/transformers.py
@@ -0,0 +1,18 @@
+import os
+
+# Don't print "None of PyTorch, TensorFlow >= 2.0, or Flax have been
+# found. Models won't be available and only tokenizers, configuration
+# and file/data utilities can be used" warning.  Tokenizers is all we
+# want!
+
+__var_name = "TRANSFORMERS_NO_ADVISORY_WARNINGS"
+__orig_val = os.environ.get(__var_name)
+os.environ[__var_name] = "1"
+try:
+    from transformers import AutoTokenizer  # noqa: F401
+finally:
+    if __orig_val is None:
+        os.environ.pop(__var_name)
+    else:
+        os.environ[__var_name] = __orig_val
+    del __var_name, __orig_val, os
diff --git a/src/dom_tokenizers/train.py b/src/dom_tokenizers/train.py
index 0d9aa56..92c7092 100644
--- a/src/dom_tokenizers/train.py
+++ b/src/dom_tokenizers/train.py
@@ -1,4 +1,3 @@
-import os
 import json
 import warnings
 
@@ -8,9 +7,7 @@
 from datasets import load_dataset
 from tokenizers.pre_tokenizers import PreTokenizer, WhitespaceSplit
 
-os.environ["TRANSFORMERS_NO_ADVISORY_WARNINGS"] = str(True)
-from transformers import AutoTokenizer
-
+from .internal.transformers import AutoTokenizer
 from .pre_tokenizers import DOMSnapshotPreTokenizer
 
 DEFAULT_BASE = "bert-base-uncased"