Fix typo.

Let me know if this slows things down on 2000 series and below.

comfy/model_management.py

@@ -1121,7 +1121,7 @@ def should_use_bf16(device=None, model_params=0, prioritize_performance=True, ma
 
     bf16_works = torch.cuda.is_bf16_supported()
 
-    if bf16_works or manual_cast:
+    if bf16_works and manual_cast:
         free_model_memory = maximum_vram_for_weights(device)
         if (not prioritize_performance) or model_params * 4 > free_model_memory:
             return True