diff --git a/unsloth/models/vision.py b/unsloth/models/vision.py
index efdf67a95e..fbb154ddd0 100644
--- a/unsloth/models/vision.py
+++ b/unsloth/models/vision.py
@@ -106,6 +106,7 @@ def unsloth_base_fast_generate(
     except: pass
 
     # Mixed precision autocast
+    if os.environ.get("UNSLOTH_FORCE_FLOAT32", "0") == "1": dtype = torch.float32
     with torch.inference_mode(), torch.autocast(device_type = "cuda", dtype = dtype):
         output = self._old_generate(*args, **kwargs)
     pass