[BitsandBytes] Verify if GPU is available (huggingface#30533)

Change order

src/transformers/quantizers/quantizer_bnb_4bit.py

@@ -58,6 +58,8 @@ def __init__(self, quantization_config, **kwargs):
  self.modules_to_not_convert = self.quantization_config.llm_int8_skip_modules
 
  def validate_environment(self, *args, **kwargs):
+ 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 `bitsandbytes` 8-bit quantization requires Accelerate: `pip install accelerate` "
@@ -70,9 +72,6 @@ def validate_environment(self, *args, **kwargs):
  " sure the weights are in PyTorch format."
  )
 
- if not torch.cuda.is_available():
- raise RuntimeError("No GPU found. A GPU is needed for quantization.")
-
  device_map = kwargs.get("device_map", None)
  if (
  device_map is not None

src/transformers/quantizers/quantizer_bnb_8bit.py

@@ -58,6 +58,9 @@ def __init__(self, quantization_config, **kwargs):
  self.modules_to_not_convert = self.quantization_config.llm_int8_skip_modules
 
  def validate_environment(self, *args, **kwargs):
+ 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 `bitsandbytes` 8-bit quantization requires Accelerate: `pip install accelerate` "
@@ -70,9 +73,6 @@ def validate_environment(self, *args, **kwargs):
  " sure the weights are in PyTorch format."
  )
 
- if not torch.cuda.is_available():
- raise RuntimeError("No GPU found. A GPU is needed for quantization.")
-
  device_map = kwargs.get("device_map", None)
  if (
  device_map is not None