diff --git a/unsloth/models/_utils.py b/unsloth/models/_utils.py
index 7af1c336fc..b047781e11 100644
--- a/unsloth/models/_utils.py
+++ b/unsloth/models/_utils.py
@@ -250,7 +250,7 @@ def determine_attention_implementation(model_class, config):
                 and (model_class is not None)
                 and getattr(model_class, "_supports_flex_attn", False)
             ):
-                # GPT-OSS, Mllama and Gemma3 use eager/sdpa attention during
+                # GPT-OSS, Mllama and Gemma3N use eager/sdpa attention during
                 # inference since flex attention returns incorrect results or errors out.
                 # GPT-OSS: left padding issues cause incorrect outputs.
                 # Mllama: _update_causal_mask uses make_flex_block_causal_mask which
diff --git a/unsloth/models/vision.py b/unsloth/models/vision.py
index c44a77eff0..210f298e65 100644
--- a/unsloth/models/vision.py
+++ b/unsloth/models/vision.py
@@ -623,9 +623,10 @@ class FastBaseModel:
         if not ("attn_implementation" in kwargs):
             kwargs["attn_implementation"] = attn_impl
         if not supports_sdpa and kwargs.get("attn_implementation") == "sdpa":
-            print(
-                f"Unsloth: {model_type_arch.title()} does not support SDPA - switching to fast eager."
-            )
+            if os.environ.get("UNSLOTH_ENABLE_FLEX_ATTENTION", "1") == "0":
+                print(
+                    f"Unsloth: {model_type_arch.title()} does not support SDPA - switching to fast eager."
+                )
             del kwargs["attn_implementation"]
 
         bnb_config = None