diff --git a/pyproject.toml b/pyproject.toml
index fb63f89ae6..1d46c8824b 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -59,7 +59,7 @@ huggingfacenotorch = [
 ]
 huggingface = [
     "unsloth[huggingfacenotorch]",
-    "unsloth_zoo>=2025.10.6",
+    "unsloth_zoo>=2025.10.8",
     "torchvision",
     "unsloth[triton]",
 ]
@@ -461,7 +461,7 @@ colab-ampere-torch220 = [
     "flash-attn>=2.6.3 ; ('linux' in sys_platform)",
 ]
 colab-new = [
-    "unsloth_zoo>=2025.10.6",
+    "unsloth_zoo>=2025.10.8",
     "packaging",
     "tyro",
     "transformers>=4.51.3,!=4.52.0,!=4.52.1,!=4.52.2,!=4.52.3,!=4.53.0,!=4.54.0,!=4.55.0,!=4.55.1,<=4.56.2",
diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index 78ca68b449..22ce2fd9e5 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -12,7 +12,7 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-__version__ = "2025.10.6"
+__version__ = "2025.10.7"
 
 __all__ = [
     "SUPPORTS_BFLOAT16",
@@ -163,7 +163,6 @@
 warnings.filterwarnings(action = "ignore", category = RuntimeWarning, module = "multiprocessing")
 warnings.filterwarnings(action = "ignore", category = RuntimeWarning, module = "multiprocess")
 warnings.filterwarnings(action = "ignore", category = UserWarning,    module = "triton")
-
 # Stop "Special tokens have been added in the vocabulary, ..."
 import logging
 logging.getLogger("transformers.tokenization_utils_base").setLevel(logging.CRITICAL+1)
@@ -215,6 +214,12 @@ def filter(self, x): return not (self.text in x.getMessage())
         del vllm_lora_model_logger
     except:
         pass
+    try:
+        from vllm.attention.utils.fa_utils import logger as vllm_attention_utils_fa_utils_logger
+        vllm_attention_utils_fa_utils_logger.addFilter(HideLoggingMessage("Cannot use FA version"))
+        del vllm_attention_utils_fa_utils_logger
+    except:
+        pass
 pass
 
 # The speedups for torchdynamo mostly come with GPU Ampere or higher and which is not detected here.
@@ -344,6 +349,14 @@ def filter(self, x): return not (self.text in x.getMessage())
 except:
     pass
 
+# We detected that you are using `from_pretrained` with a meta device context manager or `torch.set_default_device('meta')
+try:
+    from transformers.modeling_utils import logger as modeling_utils_logger
+    modeling_utils_logger.addFilter(HideLoggingMessage("anti-pattern"))
+    del modeling_utils_logger
+except:
+    pass
+
 # Errors out on
 # Some weights of Gemma3nForConditionalGeneration were not initialized from the model checkpoint
 from transformers.modeling_utils import logger as transformers_logger
diff --git a/unsloth/tokenizer_utils.py b/unsloth/tokenizer_utils.py
index 067f2596c6..b35c1326ca 100644
--- a/unsloth/tokenizer_utils.py
+++ b/unsloth/tokenizer_utils.py
@@ -345,7 +345,23 @@ def fix_sentencepiece_tokenizer(
 ):
     # From https://github.com/google/sentencepiece/issues/121
     # We need to manually edit the sentencepiece tokenizer!
-    from transformers.utils import sentencepiece_model_pb2
+    try:
+        from transformers.convert_slow_tokenizer import import_protobuf
+        sentencepiece_model_pb2 = import_protobuf()
+    except Exception as e:
+        try:
+            import google.protobuf
+            from unsloth_zoo.utils import Version
+            protobuf_version = Version(google.protobuf.__version__)
+            if protobuf_version > Version("3.20.3"):
+                raise RuntimeError(
+                    f"Unsloth: Your protobuf version = {protobuf_version} is too new.\n"\
+                    f"Please downgrade via `pip install --force-reinstall protobuf==3.20.3`"
+                )
+        except:
+            # This will only work for older SentencePiece versions <= 3.20.3
+            from transformers.utils import sentencepiece_model_pb2
+    pass
 
     if not os.path.exists(temporary_location):
         os.makedirs(temporary_location)