default detect_anomaly to True

default detect_anomaly to True as NaN or Inf losses usually don't make any sense

Author: romeokienzler

src/lightning/pytorch/trainer/trainer.py

@@ -124,7 +124,7 @@ def __init__(
         inference_mode: bool = True,
         use_distributed_sampler: bool = True,
         profiler: Optional[Union[Profiler, str]] = None,
-        detect_anomaly: bool = False,
+        detect_anomaly: bool = True,
         barebones: bool = False,
         plugins: Optional[Union[_PLUGIN_INPUT, list[_PLUGIN_INPUT]]] = None,
         sync_batchnorm: bool = False,