fix FSDP version related issues (#22489)

fix fsdp

src/transformers/trainer.py

@@ -1481,6 +1481,11 @@ class Trainer:
                     mixed_precision_policy = MixedPrecision(param_dtype=dtype, reduce_dtype=dtype, buffer_dtype=dtype)
                 if type(model) != FSDP:
                     # XXX: Breaking the self.model convention but I see no way around it for now.
+                    signature = inspect.signature(FSDP.__init__).parameters.keys()
+                    kwargs = {}
+                    for arg in ["limit_all_gathers", "forward_prefetch", "backward_prefetch"]:
+                        if arg in signature:
+                            kwargs[arg] = getattr(self, arg)
                     self.model = model = FSDP(
                         model,
                         sharding_strategy=self.fsdp,
@@ -1488,9 +1493,7 @@ class Trainer:
                         auto_wrap_policy=auto_wrap_policy,
                         mixed_precision=mixed_precision_policy,
                         device_id=self.args.device,
-                        backward_prefetch=self.backward_prefetch,
-                        forward_prefetch=self.forword_prefetch,
-                        limit_all_gathers=self.limit_all_gathers,
+                        **kwargs,
                     )
             else:
                 try: