Better error message for bitsandbytes import (#27764)

* better error message

* fix logic

* fix log

src/transformers/modeling_utils.py

@@ -1010,7 +1010,7 @@ def num_parameters(self, only_trainable: bool = False, exclude_embeddings: bool
             else:
                 raise ValueError(
                     "bitsandbytes is not installed but it seems that the model has been loaded in 4bit precision, something went wrong"
-                    " make sure to install bitsandbytes with `pip install bitsandbytes`."
+                    " make sure to install bitsandbytes with `pip install bitsandbytes`. You also need a GPU. "
                 )
 
         for param in total_parameters:
@@ -2746,11 +2746,13 @@ def from_pretrained(
                 )
 
         if load_in_8bit or load_in_4bit:
+            if not torch.cuda.is_available():
+                raise RuntimeError("No GPU found. A GPU is needed for quantization.")
             if not (is_accelerate_available() and is_bitsandbytes_available()):
                 raise ImportError(
                     "Using `load_in_8bit=True` requires Accelerate: `pip install accelerate` and the latest version of"
                     " bitsandbytes `pip install -i https://test.pypi.org/simple/ bitsandbytes` or"
-                    " pip install bitsandbytes` "
+                    " `pip install bitsandbytes`."
                 )
 
             if torch_dtype is None:
@@ -2764,10 +2766,7 @@ def from_pretrained(
                 torch_dtype = torch.float16
 
             if device_map is None:
-                if torch.cuda.is_available():
-                    device_map = {"": torch.cuda.current_device()}
-                else:
-                    raise RuntimeError("No GPU found. A GPU is needed for quantization.")
+                device_map = {"": torch.cuda.current_device()}
                 logger.info(
                     "The device_map was not initialized. "
                     "Setting device_map to {'':torch.cuda.current_device()}. "