Remove cache_enabled parameter from autocast context in load_infinity function for cleaner implementation

app.py

@@ -212,7 +212,7 @@ def load_infinity(
     text_maxlen = 512
     torch.cuda.empty_cache()
 
-    with torch.amp.autocast(device_type=device, enabled=bf16, dtype=autocast_dtype, cache_enabled=True), torch.no_grad():
+    with torch.amp.autocast(device_type=device, dtype=autocast_dtype), torch.no_grad():
         infinity_test: Infinity = Infinity(
             vae_local=vae, text_channels=text_channels, text_maxlen=text_maxlen,
             shared_aln=True, raw_scale_schedule=scale_schedule,