make new weight on cpu

Signed-off-by: Kyle Sayers <[email protected]>

Author: kylesayrs

src/llmcompressor/modeling/fuse.py

@@ -51,7 +51,8 @@ def fuse_norm_linears(norm: torch.nn.Module, linears: Iterable[torch.nn.Linear])
 
             update_offload_parameter(linear, "weight", new_weight)
 
-        update_offload_parameter(norm, "weight", torch.ones_like(norm.weight))
+        new_norm_weight = torch.ones_like(norm.weight, device="cpu")
+        update_offload_parameter(norm, "weight", new_norm_weight)
 
     else:
         raise ValueError(f"Cannot fuse norm of type {type(norm)}")