Params4bit: don't try to quantize when moving to meta device

Author: matthewdouglas

bitsandbytes/nn/modules.py

@@ -332,7 +332,7 @@ def to(self: T, tensor: Tensor, non_blocking: bool = ...) -> T: ...
     def to(self, *args, **kwargs):
         device, dtype, non_blocking, convert_to_format = torch._C._nn._parse_to(*args, **kwargs)
 
-        if device is not None and not self.bnb_quantized:
+        if device is not None and device.type != "meta" and not self.bnb_quantized:
             return self._quantize(device)
         else:
             if self.quant_state is not None: